Pub The Minds New Science A History of The Cognitive R

The Mind's
New Science
Other Books by Howard Gardner
The Quest for Mind (1973; second edition, 1981)

The Arts and Human Development (1973)
The Shattered Mind (1975)
Developmental Psychology (1978; second edition, 1982)
s
Artful Scribbles: The Significance of Children Drawings (1980)
Art, Mind, and Brain: A Cognitive Approach to Creativity (1982)
Frames of Mind: The Theory of Multiple Intelligences (1983)
THE MIND'S
NEW
SCIENCE
A History of the
Cognitive Revolution
HOWARD GARDNER
Basic Books, Inc., Publishers I New York

Quotes on pp. 66 and 88 from H. Putnam, Mind, IAnguJ~ge, Rnd Rm/ify: PhilosophiCRI PRpers,
uol. 2, 1975. Reprinted with permission, Cambridge University Press.
Quotes on pp. 2.0, 2.4, 132., and 2.95 from N. Wiener, Cybernetics, or Control Rnd CommunicRfion
in the Anima/and the Machine, 2.nd ed., 1¢11l948. Reprinted with permission, MIT Press.
Quote on p. 2.44 from Jonathan Miller, Stales of Mind, 1983. Reprinted with permission,
Pantheon Books, a Division of Random House, Inc.
Quotes on pp. 72., 73, 74, 75, and 85 from Richard Rorty, Philosophy and the Mirror of Nature.
Copyright © 1979 by Princeton University Press. Excerpts reprinted with permission of
Princeton University Press.
Quote on p. 70 from Jerome Bruner, In Sttlrck of Mind (in press). Reprinted with permission
of the author.
Library of Congress Cataloging in Publication Data

Gardner, Howard.
The mind's new science.
Bibliography: p. 393.
Includes index.
1. Cognition-History. 2.. Cognition-Research-
Methodology-History. 3· Artificial intelligence-
History. I. Title.
BF311.G339 1985 153 85-47555
ISBN ~-465-04634-7
Copyright © 1985 by Howard Gardner

Printed in the United States of America
Designed by Vincent Torre
85 86 87 88 He 9 8 7 6 5 4 3 z 1
For my parents,
Hilde Weilheimer Gardner and Ralph Gardner
CONTENTS
PREFACE .riii
PART I
THE COGNITIVE REVOLUTION
1 Introduction: What the Meno Wrought J
The Greek Agenda J
DefiniHon and Scope of CogniH'IJI! Science 5
Purpose and Plan of This Book 7
2 Laying the Foundation for Cognitive Science 10

The Hixon Symposium and the Challenge to Behaviorism 10
A Critical Moment in Scientific History 14
Key Theoretical Inputs to Cognitive Science 16
Mathtmttlics ami CompufllHon 16
The Neuronal Model 18
The CyberneHc Synthesis 19
lnformaHon Theory 21
Neuropsychological Syndromes 22
Catalytic Encounters and Influential Writings 2J
3 Cognitive Science: The First Decades 28

A Consensual Birthdate 28
The 1960s: Picking Up Steam J2
The Sloan Initiative J5
Key Features of Cognitive Science J8
RepresenfllHons J8
Computers 40
De-EmphllSis on Affect, Context, Culture, and HIStory 41
Belief in Interdisciplinary Studies 42
Rootedness in ClllSSical Philosophical Problems 42
vii
Contents
PART II
THE COGNITIVE SCIENCES: A HISTORICAL
PERSPECTIVE
4 Reason, Experience, and the Status of Philosophy 49

Descartes's Mind 49
Empiricist Responses to Descartes 54
Loclce 's Competing Model 54
Berlctley's and Humes Skeptical Tones 55
Kant and Foundational Philosophy 56
The Logical-Empiricist Program 60
The Unraveling of Logical Empiricism and the Revised Roles
of Philosophy 65
Gilbert Ryle 66
Ludwig Witfgenstein 68
f. L. Austin 69
Richard Rorty: Is Epistemology Necessary? 71
Preserving Philosophy's Purview 76
Fresh Approaches to Epistemology 78
Functionalism 78
Intentional Systems 79
The Complete Cognitivist: Jerry Fodor 81
Conclusion: The Dialectic Role of Philosophy 86
5 Psychology: The Wedding of Methods to

Substance 89
Three Pivotal Lines of Research from the 1950s 89
George Miller's Magic Number 7 89
The British ApPT'Oilch to the Processing of Information 91
Jerome Bruner's Strategic Approach 93
The Program of Cogniti'IJI! Psychology 95
Scientific Psychology in the Nineteenth Century 98
Coping with the Kantian Legacy 98
Laying the Groundwork: Helmholtz, F«hner, Donders, and
Brentano 99
Wundt's Program 102
Inno'Oiltive Methods: Hermann Ebbinghaus 104
The Early Twentieth Century 105
The Attaclc on the Wundtilms 105
Functionalism: William fames 107
The Behaviorist Revolution 109
Gestalt Psychology: A View from Above 111
Origins 111
Kohler's ComP"hensive Researches 112
viii
Contents
Frederic Bartlett's Schematic Approach 114
Jean Piaget 's Developmental Concerns 116
The Tum to Cognition 118
lnspiratitm from Computers 118
Reactions to the Standard Information-Processing Paradigms:
The Top-Down Perspective 124
Mental Representations 128
Psychology's Contributions 130
6 Artificial Intelligence: The Expert Tool 138

The Summer of 1956 at Dartmouth 138
The Ideas of Artificial Intelligence 140
The Dream of Artificial Intelligence 142
Realizing the Dream 144
The Programs of the Dartmouth Tetrad 145
Programs for Problems: Allen Newell and Herbert Simon 145
Marvin Minsky and Hts Students 151
Lists and Logics: John McCarthy 154
Other Programming Milestones 155
The Phenomenal SHRDLU 158
Pivotal Issues 160
The Need for Expert Systems 160
Procedural versus Declarative Representation 161
The Three Sharpest Cuts 162
Innovations in the 1970s 165
Pluralisms 166
Understanding of Language 167
Perceptitm 169
The Chinese Room 171
Searle's Conundrum 172
Counterattacks 173
Critics and Defenders: The Debate Continues 177
7 Linguistics: The Search for Autonomy 182

At First, There Were Colorless Green Ideas . . . 182
Enigmatic Sentences 182
Chomsky's Approach 185
The Spreading of Green Ideas 189
Verbal Mtsbehavior: The Controversy with Skinner 191
General Messages: Chomskian Themata 193
Linguistics of an Earlier Era 196
The Neo-Grammarians 197
de Saussure's Signal Ctmfributions 198
The Prague School 200
Bloomfield Fashions a Fteld 202
The Crisis of the Early 1950s 205
The Evolution of Chomsky's Thought 207
Contents
Period Pieces 207
Reactions in Other Cognitive Sciences 214
Rival Positions within Linguistics 216
A Tentative Evaluation 218
8 Anthropology: Beyond the Individual Case 223

Lucien Levy-Bruhl Examines the Mind of the Primitive 223
Edward Tylor Launches the Discipline of Anthropology 227
The British Scene 229
The American Version 231
Boas's Scholarly Hegemony 231
Reactions to Boas 233
The Special Status of Language and Linguistics 234
The Structuralist Version 236
Livi-Strauss's Canons 236
Exploring Mind 238
Myth Making 239
Sperber's VariRtions 242
Ethnoscience 244
Roots 244
A Sample Componential Analysis 246
Critiques of Ethnoscience: From Within 249
Critiques of Ethnoscience: Outside the Ranlrs 250
Psychological Forays 253
Levy-Bruhl Revisited 257
9 Neuroscience: The Flirtation with Reductionism 260

Karl Lashley Poses a Research Agenda 260
The Lesion Technique 261
Equipotentiality and Engrams 262
Lashley's Iconoclasm 263
How Specific Is Neural Functioning? 265
Euidence for Localization 267
The Resurgence of Holism 268
Evaluating the Euidence 2 70
Donald Hebb's Bold Synthesis 271
The Hixon Symposium Revisited 272
Hubel and Wiesel's Decisive Demonstrations 273
The Molar Perspective 2 74
Sperry on Split Brains 275
Gradients of Plasfidty and Hierarchy of Functions 2 76
The Neural Base of Cognition: Studies of Two Systems 278
Eric Kandel Bridges a Gap 2 79
The Song of Birds 280
Pribram 's Holographic Hypothesis 282
Three HIStorical Moments 2 84
Will Neuroscience Devour Cognitive Science? 285
X
Contents
PART III
TOWARD AN INTEGRATED COGNITIVE
SCIENCE: PRESENT EFFORTS,
FUTURE PROSPECTS
Introduction 291
10 Perceiving the World 295

Perennial Puzzles of Perception 295
Computer Simulations 297
The Work of David Marr 298
Levels of Scene Analysis 299
Two Sketches and a Model 301
Implications f(Jf Cognitive Science 305
Reactions to Marr 307
The Gibsonian View of Perception 308
Cognitive-Science Critiques of Gibson 311
An Aggressive Defense 314
Contrasting Perspectives 316
Possible Reconciliations 318
Neisser 's Ecological Approach 318
Parallel Processing in Perception 318
11 Mental Imagery: A Figment of the Imagination? 323

Introduction: Images through the Ages 323
Stephen Kosslyn's Model 326
Computer Simulation 328
The Debate about the Kosslyn-Shepard Perspective 330
Pylyshyn 's Penetrating CllSe against Imagery 332
A Wittgensteinian Critique 336
12 A World Categorized 340

The Classical View of Classification 341
The Universe of Color Terms 342
Rosch 5 Critique of Classical Views 344
Berlin and Kay on Basic Color Terminology 348
A New Philosophical Cast on Concepts 350
Can Categorization Be Studied from a Cognitivist
Perspective? 355
13 How Rational a Being? 360

The Illogic of Human Reasoning 361
xi
Contents
Cards with Numbers 361
Artists and Beekeepers 363
Mental Models as a Panacea? 367
Biases in Human Cognition: The Tversky-Kahneman
Position 370
Theater Tickets and Coin Tossers 371
A Philosophical Critique 3 73
Conclusion 3 79
14 Conclusion: The Computational Paradox and the

Cognitive Challenge 381
The Centrality of Mental Representation 383
The Computational Paradox 384
The Cognitive Challenge 389
REFERENCES 393
NAME INDEX 409
SUBJECT INDEX 415
Xtt
PREFACE
In the mid-1970s, I began to hear the term cognilive science. As a psychologist

interested in cognitive matters, I naturally became curious about the meth-
ods and scope of this new science. When I was unable to find anything
systematic written on the subject, and inquiries to colleagues left me con-
fused, I decided to probe further. Some immersion in the writings of
self-proclaimed cognitive scientists convinced me that cognitive science
was deeply rooted in philosophy and therefore, in a sense, had a long
history. At the same time, the field was so new that its leading figures were
all alive, and some of them were still quite young.
I decided that it would be useful and rewarding to undertake a study
in which I would rely heavily on the testimony of those scholars who had
founded the field as well as those who were at present its most active
workers. But in lieu of an oral history or a journalistic account of current
laboratory work (both of which subsequently were undertaken by other
authors), I decided to make a comprehensive investigation of cognitive
science in which I could include the long view-the philosophical origins,
the histories of each of the respective fields, the current work that appears
most central, and my own assessment of the prospects for this ambitious
field.
It had not escaped my attention that the Alfred P. Sloan Foundation
was a major supporter of work in the cognitive sciences. I therefore ap-
proached its program officer Kenneth Klivington about the possibility of
writing a history of cognitive science. To my delight, the Foundation
proved receptive, and I began my formal study at the beginning of 1981.
I want to express my gratitude to the entire administration of the Sloan
Foundation, and to its two responsible program officers, Kenneth Kliving-
ton and Eric Wanner, who were totally supportive of my efforts to carry
through this somewhat risky undertaking.
In the course of my study, I interviewed formally, or conducted infor-
mal discussions with, dozens of cognitive scientists in this country and
xm
Preface
abroad. As far as I can recall, no scientist whom I approached denied me
an interview, and most--even those who expressed skepticism about cog-
nitive science-were gracious and informative. I regret that I had to stop
interviewing and begin writing after a time, and I regret even more that
I ultimately was not able to discuss in print the work of many of those from
whom I learned much. Unfortunately, if I had included even half of the
work worthy of review, this book would be several times longer than
it is.
I want to mention first and thank the many individuals who willingly
discussed their work and the field of cognitive science with me. (I also must
apologize to those whom I have inadvertently omitted from this list.) I am
indebted to Jonathan Adler, Allan Allport, John Anderson, Dana Ballard,
Jon Barwise, Elizabeth Bates, Brent Berlin, Ned Block, Daniel Bobrow,
Margaret Boden, Stanley Brandes, Joan Bresnan, John Seely Brown, Roger
Brown, Jerome Bruner, Peter Bryant, Alfonso Caramazza, Noam Chomsky,
Gillian Cohen, Michael Cole, Roy D'Andrade, Daniel Dennett, Hubert
Dreyfus, Jerome Feldman, Charles Fillmore, Jerry Fodor, Michael Gaz-
zaniga, Clifford Geertz, my late and beloved mentor Norman Geschwind,
Samuel Glucksberg, Nelson Goodman, Charles Gross, Patrick Hayes,
Geoffrey Hinton, Stephen Isard, Philip Johnson-Laird, Ronald Kaplan,
Paul Kay, Samuel Jay Keyser, Stephen Kosslyn, George Lakoff, Jean Lave,
Jerome Lettvin, Robert LeVine, Claude Levi-Strauss, Christopher Longuet-
Higgins, John McCarthy, Jay McClelland, Jean Mandler, Alexander Mar-
shack, John Marshall, Jacques Mehler, Susanna Millar, George Miller,
Marvin Minsky, Julius Moravcsik, John Morton, Ulric Neisser, Freda
Newcombe, Allen Newell, Donald Norman, Daniel Osherson, Domenico
Parisi, Stanley Peters, Michael Posner, Karl Pribram, Hilary Putnam, Raj
Reddy, Richard Rorty, Eleanor Rosch, David Rumelhart, Roger Schank,
Israel Scheffler, John Searle, Robert Siegler, Herbert Simon, Aaron Sloman,
Brian Cantwell Smith, Stuart Sutherland, Leonard Talmy, Sheldon
Wagner, Terry Winograd, and Edgar Zurif.
Several friends and colleagues were good enough to read and comment
critically on one or more of the drafts of this book. I am considerably in
their debt. I wish to thank Margaret Boden, Hiram Brownell, Daniel Den-
nett, Martha Farah, Josef Grodzinsky, Jerome Kagan, Benny Shanon, Eric
Wanner, my wife, Ellen Winner, and several anonymous reviewers for
their useful comments, criticisms, and words of encouragement. I know
that I benefited greatly from their feedback; I fear that remaining errors and
infelicities are my own responsibility.
Over the several years in which this book was in preparation, I was
fortunate enough to have the help of Linda Levine, Susan McConnell,
Christine Meyer, and Claudia Strauss, who served as research assistants.
xiv
Preface
Mara Krechevsky, my current research assistant, has been invaluable in

helping me to bring the manuscript to publication. In addition, she has
made many substantive contributions to the manuscript. I thank Connie
Wolf at Harvard and Carmella Loffredo at the Sloan Foundation for their
help. The manuscript in its various guises was ably typed and word-
processed by Dolly Appel, Damaris Chapin, Isabel Eccles, Nan Kortz, and
Laura Stephens-Swannie. I am sure they would agree with the sentiment
expressed by Samuel Johnson with respect to Paradise Lost: "No man could
wish it longer."
As with my last three books, I have been fortunate to have the support
of many individuals at Basic Books. On the editorial side I am tremen-
dously grateful to Judith Greissman, Jane Isay, and Martin Kessler for their
thoughtful reactions to earlier versions of this manuscript. Linda Carbone
performed ably as the project editor; and Phoebe Hoss, as development
editor, helped me to deal with many expositional problems and also dis-
played an uncanny sense of where I (and, at times, where cognitive science)
had fallen short. In another life, she is at risk of becoming a cognitivist
herself.
My greatest pleasure is to have the opportunity to dedicate this book
to my parents.
HowARD GARDNER
Cambridge, Massachusetts
Apri/1985
XV
PART I
THE COGNITIVE
REVOLUTION
1
Introduction:
What the Meno
Wrought
One thing I would fight for to the end, both in word and
deed if I were able-that if we believed that we must try
to find out what is not known, we should be better and
braver and less idle than if we believed that what we do
not know it is impossible to find out and that we need not
even try.
-SOCRATES, The Meno
The safest general characterization of the European philo-

sophical tradition is that it consists in a series of footnotes
to Plato.
-ALFRED NoRTH WHITEHEAD
The Greek Agenda

In the Meno, a Platonic dialogue, Socrates persistently questions a
young slave about his knowledge of geometry. At first the slave appears
quite knowledgeable, readily asserting that a square composed of sides two
feet in length contains four square feet. But when, in response to a problem
posed by Socrates, the slave indicates that a figure of eight square feet
contains sides four feet long, Socrates demonstrates that the boy is
thoroughly confused and does not realize that the length of the side must
be the square root of eight.
3
I I THE COGNITIVE REVOLUTION
The centerpiece of the dialogue features many questions and re-

sponses in the approved Socratic manner. Through this interchange, the
philosopher ultimately succeeds in drawing out from the boy the knowl-
edge that a square with a four-foot side would actually be sixteen square
feet-that is, twice as great an area than he had supposed; and the knowl-
edge that one can, by geometric maneuvers, inscribe a square that is actu-
ally eight square feet within this larger square. In so doing, Socrates has
demonstrated to his satisfaction, and to the satisfaction of the slave's
master, Menon, that the youth possesses within him all of the knowledge
necessary to compute the various geometrical relationships in question.
At issue in this Platonic dialogue was far more than an exploration of
the extent of knowledge possessed by a single slave boy. Here, for perhaps
the first time in human intellectual history, was an extended rumination
on the nature of knowledge: where does it come from, what does it consist
of, how is it represented in the human mind? And, for good measure, there
was also proposed a specific-if ultimately highly controversial-theory of
human knowledge.
According to Plato (and, presumably, Socrates as well), the domain of
knowledge par excellence inhered in mathematics and the exact sciences
it had spawned. Indeed, the purest forms of knowledge were idealized
forms or archetypes which can merely be glimpsed in mundane reality. An
understanding of geometrical matters-indeed, of all matters of genuine
knowledge-was already implanted in the human soul at birth. The task
in instruction, as demonstrated in the dialogue of the Meno, was simply to
bring this innate knowledge to conscious awareness.
The Greeks' interest in the nature of knowledge, no less than their
particular contentious theories and evocative images, continued to rever-
berate through the Western intellectual tradition. Aristotle's version was
the principal cornerstone of the Middle Ages, when discussions about
knowledge were principally the purview of theologians. Then, during the
Renaissance and Enlightenment periods, philosophers continued the dis-
cussions and began to draw regularly on findings obtained in the newly
emerging empirical sciences. Such thinkers as Descartes, Locke, and Kant
dealt comfortably with theoretical and empirical issues concerning knowl-
edge, and the Neapolitan scholar Giambattista Vico even christened a New
Science (Scienza Nova) to deal with these and related matters. By the end of
the nineteenth century, there had been a proliferation of new sciences and
philosophical specialties, several of which purported to deal with the na-
ture of the human mind.
Today, armed with tools and concepts unimaginable even a century
ago, a new cadre of thinkers called cognitive scientists has been investigat-
ing many of the same issues that first possessed the Greeks some twenty-
4
Introduction: What the Meno Wrought
five hundred years ago. Like their earlier counterparts, cognitive scientists
today ask what it means to know something and to have accurate beliefs,
or to be ignorant or mistaken. They seek to understand what is known-
the objects and subjects in the external world-and the person who knows
-his" perceptual apparatus, mechanisms of learning, memory, and ratio-
nality. They ponder the sources of knowledge: where does it come from,
how is it stored and tapped, how might it be lost? They are curious about
the differences among individuals: who learns early or with difficulty;
what can be known by the child, the inhabitant of a preliterate society, an
individual who has suffered brain damage, or a mature scientist?
Further, cognitive scientists, again as did the Greeks, conjecture about
the various vehicles of knowledge: what is a form, an image, a concept, a
11
word; and how do these modes of representation" relate to one another?
They wonder about the priorities of specific sense organs as against a
central"general understanding" or "common sense." They reflect on lan-
guage, noting the power and traps entailed in the use of words and their
possible predominant influence over thoughts and beliefs. And they spec-
ulate at length on the nature of the very activity of knowing: why do we
want to know, what are the constraints on knowing, and what are the
limits of scientific knowledge about human knowing?
This "new science," thus, reaches back to the Greeks in the commit-
ment of its members to unraveling the nature of human knowledge. At the
same time, however, it is radically new. Proceeding well beyond armchair
speculation, cognitive scientists are fully wedded to the use of empirical
methods for testing their theories and their hypotheses, of making them
susceptible to disconfirmation. Their guiding questions are not just a re-
hash of the Greek agenda: new disciplines, like artificial intelligence, have
arisen; and new questions, like the potential of man-made devices to think,
stimulate research. Moreover, cognitive scientists embrace the most recent
scientific and technological breakthroughs in a variety of disciplines. Most
central to their undertaking is the computer-that creation of the mid-
twentieth century that holds promise for changing our conceptions of the
world in which we live and our picture of the human mind.
Definition and Scope of Cognitive Science

In the course of proposing and founding a new field of knowledge,
many individuals will formulate their own definitions. Indeed, since the
term cognilive science first began to be bandied about in the early 1970s,
dozens of scientists have attempted to define the nature and scope of the
*For ease of exposition, the pronoun he is used in its generic sense throughout this book.
5
field (see, for example, Bruner 1983; Collins 1977; Mandler 1981; Miller
1979; Norman 1980; Rumelhart 1982). It therefore becomes important for
me at the outset to state what I take cognitive science to be.
I define cognitive science as a contemporary, empirically based effort
to answer long-standing epistemological questions-particularly those
concerned with the nature of knowledge, its components, its sources, its
development, and its deployment. Though the term cognilive sdence is some-
times extended to include all forms of knowledge-animate as well as
inanimate, human as well as nonhuman-! apply the term chiefly to efforts
to explain human knowledge. I am interested in whether questions that
intrigued our philosophical ancestors can be decisively answered, instruc-
tively reformulated, or permanently scuttled. Today cognitive science
holds the key to whether they can be.
Of the various features or aspects generally associated with cognitive-
scientific efforts, I consider five to be of paramount importance. Not every
cognitive scientist embraces every feature, of course; but these features can
be considered symptomatic of the cognitive-scientific enterprise. When all
or most are present, one can assume that one is dealing with cognitive
science; when few, if any, are present, one has fallen outside my definition
of cognitive science. These features will be introduced more formally at the
end of chapter 3 and will be revisited repeatedly throughout the book, but
it is important to make an initial acquaintance with them at this point.
First of all, there is the belief that, in talking about human cognitive
activities, it is necessary to speak about mental representations and to posit
a level of analysis wholly separate from the biological or neurological, on
the one hand, and the sociological or cultural, on the other.
Second, there is the faith that central to any understanding of the
human mind is the electronic computer. Not only are computers indispens-
able for carrying out studies of various sorts, but, more crucially, the
computer also serves as the most viable model of how the human mind
functions.
While the first two features incorporate the central beliefs of current
cognitive science, the latter three concern methodological or strategic char-
acteristics. The third feature of cognitive science is the deliberate decision
to de-emphasize certain factors which may be important for cognitive
functioning but whose inclusion at this point would unnecessarily compli-
cate the cognitive-scientific enterprise. These factors include the influence
of affective factors or emotions, the contribution of historical and cultural
factors, and the role of the background context in which particular actions
or thoughts occur.
As a fourth feature, cognitive scientists harbor the faith that much is
to be gained from interdisciplinary studies. At present most cognitive
6
scientists are drawn from the ranks of specific disciplines-in particular,
philosophy, psychology, artificial intelligence, linguistics, anthropology,
and neuroscience (I shall refer to these disciplines severally as the "cogni-
tive sciences"). The hope is that some day the boundaries between these
disciplines may become attenuated or perhaps disappear altogether, yield-
ing a single, unified cognitive science.
A fifth and somewhat more controversial feature is the claim that a
key ingredient in contemporary cognitive science is the agenda of issues,
and the set of concerns, which have long exercised epistemologists in the
Western philosophical tradition. To my mind, it is virtually unthinkable
that cognitive science would exist, let alone assume its current form, had
there not been a philosophical tradition dating back to the time of the
Greeks.
Purpose and Plan of This Book

I have chosen to write a book on cognitive science because I consider
this area to be the most exciting new line of inquiry undertaken by scien-
tists in the past few decades. Whether it will ultimately achieve all of its
objectives, no one can say at this pointi but this seems an opportune time
to present a history and a current assessment. For contemporaries present
during the opening decades of cognitive science, I hope to convey some-
thing of the enthusiasm I have noted, the difficulties that are being con-
fronted, and the nature of the research enterprises in which investigators
are presently engaged.
My history has two components. The first consists of the various
interdisciplinary conversations and projects that took place in this century
-both those preceding and those surrounding the unofficial launching of
cognitive science in the rnid-1950s. I relate the founding of cognitive
science in the next two chapters of the book. The second component-
spanning chapters 4 through 9-consists of brief targeted histories of each
of the six aforementioned fields of cognitive science. (Other disciplines,
11
such as sociology or economics, might have been added; the borderline"
disciplines of anthropology and neuroscience might have been elirninatedi
but I believe that the major points about cognitive science are made effec-
tively by these six fields.) In my view, a brief targeted history of each of
the several cognitive sciences serves as an optimal introduction to the
principal issues of today, to the ways in which they are currently ap-
proached and explored, and to the lines of work likely to be undertaken
in the future.
I have built each historical chapter around one or two major themes,
which have been selected to convey a feeling for the kinds of issues that
7
I 1 THE CoGNITIVE REvoLUTION
have recurred and the kinds of approaches that are especially central
within a particular field. For example, in philosophy I trace the perennial
dispute between those of a rationalist persuasion (who view the mind as
actively organizing experiences on the basis of pre-existing schemes); and
those of an empiricist bent (who treat mental processes as a reflection of
information obtained from the environment). In anthropology I survey
various attempts over the years to compare the thought of primitive peo-
ples with that exhibited by typical individuals in modem Western society.
Approaching these same fields from a methodological point of view, I raise
the questions whether philosophy will eventually come to be supplanted
by an empirically based cognitive science, and whether anthropology can
(or even should) ever transcend the individual case study.
Of course, such organizing themes can only scratch the surface of the
complex territory that underlies any scientific discipline. Still I hope that
through such themes I can convey how a linguist views an issue, what a
psychologist deems a problem (and a solution), which conceptions of
process obtain in neuroscience and artificial intelligence. Only through
such an immersion in the daily (and yearly) concerns of a cognitive scien-
tist drawn from a particular discipline can one appreciate the possibilities
-and the difficulties-that arise when workers from different fields col-
laborate in cognitive-scientific research. In the end I will in each case take
stock and indicate where things stand with reference to the principal lines
of contention in a particular cognitive science-a discussion that will, in
tum, suggest some of the principal factors that have stimulated cognitive
scientists to join forces.
While each of the histories stands on its own, their juxtaposition
points up fascinating and difficult-to-anticipate parallels. Scientific fields
hardly develop in a vacuum: such disparate factors as the dissemination
of Darwin's pivotal writings, the outbreak of wars, the rise of great univer-
sities have had reverberations-and sometimes cataclysmic ones-across
apparently remote fields, which may well have had little direct contact
with one another. For the most part, I shall simply allow these parallels to
emerge, but at the beginning of part III I shall specify certain historical
forces that seem to have exerted influence across a range of cognitive
sciences.
Having taken the measure of the individual cognitive sciences, I tum
in the third part of the book to review ongoing work that is quintessen-
tially cognitive-scientific. Thus, in chapters 10 to 13, the focus shifts from
work within a traditional discipline to those lines of research that stand
most squarely at the intersection of a number of disciplines and therefore
can be considered prototypical of a single, unified cognitive science. I have
sought to identify work that is of the highest quality: if cognitive science
8
is to be assessed as an intellectual enterprise, it ought to be judged by the
most outstanding instances.
There is a common structure to these four essays on current cognitive-
scientific work. Consistent with my claim that cognitive science seeks to
elucidate basic philosophical questions, each chapter begins with a peren-
nial epistemological issue. For example, in chapter 10, I describe work on
how we perceive the world; in chapter 13, I review competing claims on
the extent of human rationality. Across chapters 10 to 13, there is a pro-
gression from those issues that seem most circumscribed to those that are
most global. Not surprisingly, the most confident answers exist for the
delimited questions, while the global topics remain ringed by unresolved
questions.
My personal reflections on cognitive science are reserved for the final
chapter. There I revisit the major themes of cognitive science in light of the
histories sketched and the interdisciplinary work reviewed. I also discuss
two themes that emerge from the inquiry and that will be introduced at
greater length in chapter 3: the computational paradox and the cognitive
challenge. In my view, the future of cognitive science rests on how the
computational paradox is resolved and on how the cognitive challenge is
met.
One might say that cognitive science has a very long past but a
relatively short history. The reason is that its roots go back to classical
times, but it has emerged as a recognized pursuit only in the last few
decades. Indeed, it seems fair to maintain that the various components that
gave rise to cognitive science were all present in the early part of the
century, and the actual birthdate occurred shortly after mid-century. Just
why cognitive science arose when it did in the form it did will constitute
my story in the remainder of part I.
9
2
Laying the
Foundation for
Cognitive Science
The Hixon Symposium and the Challenge to Behaviorism
In September of 1948 on the campus of the California Institute of Technol-

ogy, a group of eminent scientists representing several disciplines met for
a conference on "Cerebral Mechanisms in Behavior," sponsored by the
Hixon Fund (Jeffress 1951). This conference had been designed to facilitate
discussions about a classic issue: the way in which the nervous system
controls behavior. And yet the discussions ranged far more widely than the
official topic had implied. For example, the opening speaker, mathemati-
cian John von Neumann, forged a striking comparison between the elec-
tronic computer (then a discovery so new that it smacked of science fiction)
and the brain (which had been around for a while). The next speaker,
mathematician and neurophysiologist Warren McCulloch, used his pro-
vocative title ("Why the Mind Is in the Head") to launch a far-ranging
discussion on how the brain processes information-like von Neumann,
he wanted to exploit certain parallels between the nervous system and
"logical devices" in order to figure out why we perceive the world the way
we do.
Less steeped in the latest technological innovations but more versed
in the problems of explaining human behavior, the next speaker, psy-
10
Laying the Foundation for Cognitive Science
chologist Karl Lashley, gave the most iconoclastic and most memorable
address. Speaking on "The Problem of Serial Order in Behavior," he chal-
lenged the doctrine (or dogma) that had dominated psychological analysis
for the past several decades and laid out a whole new agenda for research.
In the terms of my own discussion, Lashley identified some of the major
components needed for a cognitive science, even as he castigated those
forces that had prevented its emergence before this time.
In order to appreciate the importance of Lashley's remarks, it is neces-
sary to consider the scientific climate in which he (and his numerous
colleagues interested in human psychology) had been operating during the
past few decades. At the turn of the century, in the wake of the founding
of new human sciences, investigators had been addressing the key issues
of mental life: thinking, problem solving, the nature of consciousness, the
unique aspects of human language and culture. These discussions had
linked up with the philosophical agenda of the West, but investigators had
sought to go beyond sheer speculation through the use of rigorous experi-
mental methods.
Unfortunately the scientific method favored by most researchers at
that time was introspection: self-reflection on the part of a trained observer
about the nature and course of his own thought patterns. Though sugges-
tive (indeed, often too suggestive), such introspection did not lead to that
accumulation of knowledge that is critical to science. lntrospectionism
might have collapsed of its own weight, but, in fact, it was toppled in a
more aggressive manner by a group of mostly young, mostly American
scientists who became known as the "behaviorists."
The behaviorists put forth two related propositions. First of all, those
researchers interested in a science of behavior ought to restrict themselves
strictly to public methods of observations, which any scientist could apply
and quantify. No subjective ruminations or private introspection: if a
discipline were to be science, its elements should be as observable as the
physicist's cloud chamber or the chemist's flask. Second, those interested
in a science of behavior ought to focus exclusively on behavior: researchers
ought assiduously to eschew such topics as mind, thinking, or imagination
and such concepts as plans, desires, or intentions. Nor ought they to
countenance hypothetical mental constructs like symbols, ideas, schemas,
or other possible forms of mental representation. Such constructs, never
adequately clarified by earlier philosophers, had gotten the introspection-
ists into hot water. According to behaviorists, all psychological activity can
be adequately explained without resorting to these mysterious mentalistic
entities.
A strong component of the behaviorist canon was the belief in the
supremacy and determining power of the environment. Rather than in-
11
I I THE CoGNITIVE REVOLUTION
dividuals acting as they do because of their own ideas and intentions, or

because their cognitive apparatuses embody certain autonomous structur-
ing tendencies, individuals were seen as passive reflectors of various forces
and factors in their environment. An elaborate explanatory apparatus de-
tailing principles of conditioning and reinforcement was constructed in
order to explain just how such learning and shaping of particular behavior
might come about. It was believed that the science of behavior, as fash-
ioned by scholars such as Ivan Pavlov, B. F. Skinner, E. L. Thorndike, and
J. B. Watson, could account for anything an individual might do, as well
as the circumstances under which one might do it. (What one thinks was
considered irrelevant from this perspective-unless thought was simply
redefined as covert behavior.) Just as mechanics had explained the laws of
the physical world, mechanistic models built on the reflex arc could explain
human activity.
Behaviorism spoke to many needs in the scientific community, includ-
ing some that were quite legitimate: discomfort with the acceptance of
introspective evidence on face value, without any means of scientific con-
trol or any possibility for refutation; dissatisfaction with vague and global
concepts like will or purpose and the desire to explain human behavior using
the same constructs that were applied (with apparently great success) to
animal behavior. Indeed, in the wake of the troubles that had arisen from
reliance on introspectionism (troubles that are spelled out in chapter 4),
behaviorism seemed like a breath of fresh air during the opening decades
of the century. Little wonder that it caught on quickly and captured the
best minds of a generation of workers.
Yet, in retrospect, the price paid by strict adherence to behaviorism
was far too dear. So long as behaviorism held sway-that is, during the
1920s, 1930s, and 1940~uestions about the nature of human language,
planning, problem solving, imagination, and the like could only be ap-
proached stealthily and with difficulty, if they were tolerated at all. Lash-
ley's paper crystallized a growing awareness on the part of thoughtful
scientists that adherence to behaviorist canons was making a scientific
study of mind impossible.
Lashley realized that before new insights about the brain, or about
computers, could be brought to bear in the psychological sciences, it would
be necessary to confront behaviorism directly. Therefore, in his opening
remarks, Lashley voiced his conviction that any theory of human activity
would have to account for complexly organized behaviors like playing
tennis, performing on a musical instrument, and-above all-speaking. He
commented, "The problems raised by the organization of language seem
to me to be characteristic of almost all other cerebral activity" (quoted in
Jeffress 1951, p. 121). In this one sentence, Lashley put at the very center
12
of human psychology a topic that had been relegated to obscurity by his
behaviorist colleagues. At the same time, he added, the dominant theoreti-
cal explanatory framework in neurophysiology no less than in psychology
-that of simple associative chains between a stimulus and a response-
could not possibly account for any of this serially ordered behavior. The
reason is that these action sequences unfold with such rapidity that there
is no way in which the next step in the chain can be based upon the
previous one: when one plays an arpeggio, for instance, there is simply no
time for feedback, no time for the next tone to depend upon or in any way
to reflect the course of the preceding one. Similarly, the kinds of error made
by individuals-for example, slips of the tongue-often include anticipa-
tion of words that are to occur only much later in a sequence. Again, these
phenomena defy explanations in terms of linear "A evokes B" chains.
According to Lashley, these behavioral sequences have to be planned
and organized in advance. The organization is best thought of as hierarchi-
cal: there are the broadest overall plans, within which increasingly fine-
grained sequences of actions are orchestrated. Thus, for instance, in the
case of speech, the highest nodes of the hierarchy involve the overall
intention prompting the utterance, while the choice of syntax and the
actual production of sounds occupy lower nodes of the hierarchy. The
nervous system contains an overall plan or structure, within which indi-
vidual response units can-indeed, have to-be slotted, independent of
specific feedback from the environment. Rather than behavior being con-
sequent upon environmental promptings, central brain processes actually
precede and dictate the ways in which an organism carries out complex
behavior. Or, to put it simply, Lashley concluded that the form precedes
and determines specific behavior: rather than being imposed from without,
organization emanates from within the organism.
Even as he defied the standard behavioral analysis of the time, Lashley
was also challenging two major dogmas of neurobehavioral analysis: the
belief that the nervous system is in a state of inactivity most of the time,
and the belief that isolated reflexes are activated only when specific forms
of stimulation make their appearance. Lashley's nervous system consisted
of always active, hierarchically organized units, with control emanating
from the center rather than from peripheral stimulation. As he put it,
"Attempts to express cerebral function in terms of the concepts of the
reflex arc, or of associated chains of neurons, seem to me doomed to failure
because they start with the assumption of a static nervous system. Every
bit of evidence available indicated a dynamic, constantly active system, or,
rather, a composite of many interacting systems" (quoted in Jeffress 1951,
p. 135).
In the topics he chose to address, and in the ways in which he ad-
13
I I THE CoGNITIVE REvOLUTION
dressed them, Lashley was adopting a radical position. Scientists concerned

with human behavior had been reluctant to investigate human language,
because of its complexity and its relative "invisibility" as a form of behav-
ior; and when they did treat language, they typically sought analogies to
simpler forms (like running a maze or pecking in a cage) in simpler organ-
isms (like rats or pigeons). Not only did Lashley focus on language, but he
reveled in its complexity and insisted that other motoric activities were
equally intricate.
Ordinarily, a scientist who challenges established wisdom is in for a
rough time. It is rare, at a scientific meeting, for major scholars (an ambi-
tious and often jealous lot) to pay homage to a colleague. But from com-
ments by those attending the Hixon Symposium, it seemed clear that
Lashley's colleagues were deeply impressed by the originality and bril-
liance of this presentation-coming from a scientist closely associated with
the behaviorist tradition. Lashley himself declared, "I have been rather
embarrassed by some of the flattering remarks made today" (quoted in
Jeffress 1951, p. 144). It is no exaggeration to suggest that entrenched
modes of explanation were beginning to topple and that a whole new
agenda was confronting the biological and behavioral communities.
A Critical Moment in Scientific History
The scholars in attendance at the Hixon Symposium stood at a critical

juncture of scientific history. They were keenly aware of the staggering
advances of previous centuries in the physical sciences as well as of recent
breakthroughs in the biological and neural sciences. Indeed, by the middle
of the twentieth century, two major mysteries of ancient times-the nature
of physical matter and the nature of living matter-were well on their way
to being unraveled. At the same time, however, a third mystery that had
also fascinated the ancients-the enigma of the human mind-had yet to
achieve comparable clarification.
Trained (like many scholars of their time) in the humanities as well
as in the sciences, the Hixon symposiasts displayed a familiarity with the
kinds of epistemological issue that had first exercised the Greeks and had
then formed a major part of learned conversation during the Enlighten-
ment. They knew that, in the wake of Darwin's influential account of the
origin and evolution of species, many scientists had sought to bring com-
parable rigor to the study of human behavior and thought. Often spurning
direct ties to philosophy (which they regarded as a regressive intellectual
14
Laying the FoundaHon for CogniHve Science
force), these scholars at the end of the nineteenth century had launched
separate scientific disciplines, like psychology, linguistics, anthropology,
sociology, and various neurosciences. That these aspiring scientists of
human nature had succeeded in establishing effective institutional bases
within the universities could not be disputed; but the extent to which each
new discipline had arrived at important truths was still being debated at
mid-century. Finally, those attending the Pasadena meeting were well
acquainted with the scientific program of the behaviorists. And they
shared an intuition-strongly bolstered by Lashley's tightly reasoned
paper-that the behaviorist answer to questions of the human mind was
no answer at all.
But other factors had also impeded the proper launching of a science
of cognition. Fitting comfortably with behaviorism were several philo-
sophical schools-positivism, physicalism, and verificationism-which es-
chewed entities (like concepts or ideas) that could not be readily observed
and reliably measured. There was also the intoxication with psychoanal-
ysis. While many scholars were intrigued by Freud's intuitions, they felt
that no scientific discipline could be constructed on the basis of clinical
interviews and retrospectively constructed personal histories; moreover,
they deeply resented the pretense of a field that did not leave itself suscep-
tible to disconfirmation. Between the "hard line" credo of the Establish-
ment behaviorists and the unbridled conjecturing of the Freudians, it was
difficult to focus in a scientifically respectable way on the territory of
human thought processes.
Finally, the world political situation had exerted a crippling effect on
the scientific enterprise. First, the European scientific establishment had
been ripped apart by the rise of totalitarianism, and then the American
scientific establishment had been asked to lay aside its theoretical agenda
in order to help wage the war.
While the war had been, in many ways, the worst of times, bringing
on the death or disability of many talented investigators, it had also stimu-
lated certain scientific and technological activities. Within the United
States, the war effort demanded calculating machines that could "crunch"
large sets of numbers very quickly. Computers soon became a reality.
There were other war needs to be met as well. For instance, the mathemati-
cian Norbert Wiener was asked to devise more accurate anti-aircraft ma-
chinery. This work required "a good gun, a good projectile, and a fire-
control system that enables the gunner to know the target's position, apply
corrections to the gun controls, and set the fuse properly, so that it will
detonate the projectile at the right instant" (quoted in Heims 1980, p. 183).
While working on these problems at the Massachusetts Institute of Tech-
nology, Wiener and his associate, a young engineer named Julian Bigelow,
15
concluded that there were important analogies between the feedback as-
pects of engineering devices and the homeostatic processes by which the
human nervous system sustains purposive activity. These ideas of plan-
ning, purpose, and feedback, developed with mathematical precision, were
directly antithetical to the behaviorist credo. War also produced many
victims of gunfire; and medical practitioners who cared for brain-injured
patients were being asked to evaluate which tasks could be carried out and
which ones had been compromised-temporarily or permanently-by in-
jury to the nervous system. Also, a host of more person-centered issues-
ranging from the study of the effects of propaganda to the selection of men
fit to lead combat units-enlisted the efforts of behavioral scientists and
generated ideas on which the postwar human sciences were to build
(Bruner 1944; Murray 1945; Stouffer et al. 1949). So it was in other war-
tom lands, from Alan Turing and Kenneth Craik's interest in computers
in England, to Alexander Luria's painstaking research with brain-injured
patients in Russia during the war.
By the late 1940s, there was beginning to be a feeling abroad-one
observable at Pasadena but in no way restricted to that site-that perhaps
the time was ripe for a new and finally effective scientific onslaught on the
human mind. Interestingly, nearly all of the work that came to fruition in
the postwar era was in fact built upon prior theoretical efforts-work often
dating back to the beginning of the century. But this work had sometimes
been obscured by the behaviorist movement and had sometimes been
transformed in unanticipated ways by the events of the war. These ideas,
these key inputs to contemporary efforts in cognitive science, were already
familiar to the participants at the Hixon Symposium and to other scholars
involved in the first concerted efforts to found cognitive science during the
1940s and 1950s. Now it was time to put these ideas to optimal scientific
use.
Key Theoretical Inputs to Cognitive Science
Mathematics and Computation

The years around the turn of the century were of exceptional impor-
tance in mathematics and logic. For nearly two thousand years, the logic
of syllogistic reasoning developed in classical times by Aristotle had held
sway; but thanks to the work of the German logician Gottlob Frege, a new
form of logic, which involved the manipulation of abstract symbols, began
16
to evolve toward the end of the nineteenth century. Then, in the early
twentieth century, as I shall elaborate in chapter 4, the British mathemati-
cal logicians Bertrand Russell and Alfred North Whitehead sought, with
considerable success, to reduce the basic laws of arithmetic to propositions
of elementary logic. The Whitehead-Russell work influenced a whole gen-
eration of mathematically oriented thinkers, including both Norbert Wie-
ner and John von Neumann, two of the most important contributors to the
founding of cognitive science.
In the 1930s, the logical-mathematical work that ultimately had the
greatest import for cognitive science was being carried out by a then
relatively unknown British mathematician, Alan Turing. In 1936, he devel-
oped the notion of a simple machine (subsequently dubbed a "Turing
machine") which could in principle carry out any possible conceivable
calculation. The notions underlying this "theoretical" machine were sim-
ple. All one needed was an infinitely long tape which could pass through
the machine and a scanner to read what was on the tape. The tape itself
was divided into identical squares, each of which contained upon it either
a blank or some kind of slash. The machine could carry out four moves
with the tape: move to the right, move to the left, erase the slash, or print
the slash. With just these simple operations, the machine could execute
any kind of program or plan that could be expressed in a binary code (for
example, a code of blanks and slashes). More generally, so long as one
could express clearly the steps needed to carry out a task, it could be
programmed and carried out by the Turing machine, which would simply
scan the tape (no matter what its length) and carry out the instructions
(Davis 1958; McCorduck 1979).
Turing's demonstration-and the theorem he proved-was of pro-
found importance for those researchers interested in computing devices. It
suggested that a binary code (composed simply of zeros and ones) would
make possible the devising and execution of an indefinite number of
programs, and that machines operating on this principle could be con-
structed. As Turing himself pondered computing devices, he became in-
creasingly enthusiastic about their possibilities. In fact, in 1950 (shortly
before his untimely death by suicide in his early forties) he suggested that
one could so program a machine that it would be impossible to discrimi-
nate its answers to an interlocutor from those contrived by a living human
being-a notion immortalized as the "Turing machine test." This test is
used to refute anyone who doubts that a computer can really think: if an
observer cannot distinguish the responses of a programmed machine from
those of a human being, the machine is said to have passed the Turing test
(Turing 1963).
The implications of these ideas were quickly seized upon by scientists
17
interested in human thought, who realized that if they could describe with
precision the behavior or thought processes of an organism, they might be
able to design a computing machine that operated in identical fashion. It
thus might be possible to test on a computer the plausibility of notions
about how a human being actually functions, and perhaps even to con-
struct machines about which one could confidently assert that they think
just like human beings.
In building upon Turing's ideas, John von Neumann pursued the
notion of devising a program to instruct the Turing machine to reproduce
itself. Here was the powerful idea of a stored program: that is, the computer
could be controlled through a program that itself was stored within the
computer's internal memory, so that the machine would not have to be
laboriously reprogrammed for each new task (see Goldstine 1972). For the
first time, it became conceivable that a computer might prepare and
execute its own programs.
The Neuronal Model

A second line of thinking important for those involved in founding
cognitive science was put forth during the early 1940s by Warren McCul-
loch, the second speaker at the Hixon Symposium, and Walter Pitts, a
young logician. Again, the core idea was disarmingly simple, though the
actual mathematical analysis was anything but trivial. McCulloch and Pitts
(1943) showed that the operations of a nerve cell and its connections with
other nerve cells (a so-called neural network) could be modeled in terms
of logic. Nerves could be thought of as logical statements, and the ali-or-
none property of nerves firing (or not firing) could be compared to the
operation of the propositional calculus (where a statement is either true or
false) (Heims 1980, p. 211). This model allowed one to think of a neuron
as being activated and then firing another neuron, in the same way that
an element or a proposition in a logical sequence can imply some other
proposition: thus, whether one is dealing with logic or neurons, entity A
plus entity B can imply entity C. Moreover, the analogy between neurons
and logic could be thought of in electrical terms-as signals that either
pass, or fail to pass, through a circuit. The end result of the McCulloch-
Pitts demonstration: "Anything that can be exhaustively and unambigu-
ously described ... is ... realizable by a suitable finite neural network"
(von Neumann, quoted in Bernstein 1982, p. 68).
The designers of the new computational devices were intrigued by the
ideas put forth by McCulloch and Pitts. Thanks to their demonstration, the
notion of a Turing machine now looked in two directions-toward a ner-
vous system, composed of innumerable aU-or-none neurons; and toward
18
Laying the Foundation {or Cognitive Science
a computer that could realize any process that can be unambiguously
described. Turing had demonstrated the possibility in principle of comput-
ing machines of great power, while McCulloch and Pitts had demonstrated
that at least one redoubtable machine-the human brain-could be
thought of as operating via the principles of logic and, thus, as a powerful
computer.
Ultimately, McCulloch may have carried his own chain of reasoning
too far. He was convinced that fundamental problems of epistemology
could be stated and solved only in light of the knowledge of the central
nervous system (McCorduck 1979), and he tied his claims about thinking
very closely to what was known during his own time about the nervous
system. Some commentators even feel that the search by McCulloch and
his associates for a direct mapping between logic machines and the nervous
system was a regressive element in the development of cognitive science:
rather than trying to build machines that mimic the brain at a physiological
level, analogies should have been propounded and pursued on a much
higher level-for example, between the thinlcing that goes on in human
problem solving and the strategies embodied in a computer program
(McCarthy 1984). On the other hand, it was due in part to McCulloch's
own analysis that some of the most important aspects of the nervous
system came to be better understood: for he sponsored research on the
highly specific properties of individual nerve cells. Moreover, very recently
computer scientists have once again been drawing directly on ideas about
the nature of and connections among nerve cells (see chapter 10, pp.
318-22). On balance, his polymathic spirit seems to have been a benign
catalyst for the growth of cognitive science.
The Cybernetic Synthesis

Even as John von Neumann, working at Princeton, was trying to piece
together evidence from mathematics, logic, and the nervous system, math-
ematician Norbert Wiener was engaged in parallel pursuits at the Massa-
chusetts Institute of Technology (see Heims 1980; Wiener 1964). Even
more than von Neumann, Wiener had been a mathematical prodigy and,
like his counterpart, had made fundamental discoveries in mathematics
while still in his early twenties (Wiener had worked on Brownian motion;
von Neumann, on quantum theory). Clearly, in these early choices, both
men exhibited a practical bent in their mathematics: further, they aspired
to influence the growth of science and technology within their society.
During the 1930s and 1940s, Norbert Wiener, by then ensconced at
M.I.T., became involved in a variety of worldly projects. In working on
servomechanisms-devices that kept anti-aircraft artillery, guided mis-
19
siles, and airplanes on course-he had come to think about the nature of
feedback and of self-correcting and self-regulating systems, be they me-
chanical or human. He collaborated closely with Vannevar Bush, who had
pioneered in the development of analog computers. Wiener was also struck
by the importance of the work of his sometime colleagues McCulloch and
Pitts, particularly by the suggestive analogies between a system of logical
connections and the human nervous system.
Wiener went beyond all of his contemporaries in his missionary con-
viction that these various scientific and technological developments co-
hered. Indeed, in his mind they constituted a new science-one founded
on the issues of control and communication, which he deemed to be central
in the middle of the twentieth century. He first publicly formulated this
point of view in a 1943 paper, "Behavior, Purpose, and Teleology" (Ro-
senblueth, Wiener, and Bigelow 1943), in which he and his fellow authors
put forth the notion that problems of control engineering and communica-
tion engineering are inseparable; moreover, they center not on the tech-
niques of electrical engineering, but rather on the much more fundamental
notion of the message-"whether this should be transmitted by electrical,
mechanical, or nervous means." The authors introduced a then-radical
notion: that it is legitimate to speak of machines that exhibit feedback as
"striving toward goals," as calculating the difference between their goals
and their actual performance, and as then working to reduce those differ-
ences. Machines were purposeful. The authors also developed a novel
notion of the central nervous system. As Wiener later put it:
The central nervous system no longer appears as a self-contained organ,

receiving inputs from the senses and discharging into the muscles. On the contrary,
some of its most characteristic activities are explicable only as circular processes,
emerging from the nervous system into the muscles, and re-entering the nervous
system through the sense organs, whether they be proprioceptors or organs of the
special senses. This seemed to us to mark a new step in the study of that part of
neurophysiology which concerns not solely the elementary processes of nerves and
synapses but the performance of the nervous system as an integrated whole.
(Wiener 1961, p. 8)
The parallels to Lashley's ideas about neural organization-and the chal-

lenge to behaviorist reflexology-are striking indeed.
Before long, Wiener had contrived a synthesis of the various inter-
locking ideas and presented it in the landmark volume Cybernetics (first
published in 1948, the same year as the Hixon Symposium). He introduced
his neologistic science as follows: "We have decided to call the entire field
of control and communication theory, whether in the machine or in the
animal, by the name Cybernetics" (1961, p. 11). In the following pages, he
20
Laying the FoundaHon for CogniHve Science
set down an integrated vision-a linkage of developments in understand-
ing the human nervous system, the electronic computer, and the operation
of other machines. And he underscored his belief-echoing von Neumann
and McCulloch and Pitts-that the functioning of the living organism and
the operation of the new communication machines exhibited crucial paral-
lels. Though Wiener's synthesis was not ultimately the one embraced by
cognitive science (it came closer to achieving that exalted status in the
Soviet Union), it stands as a pioneering example of the viability of such
an interdisciplinary undertaking.
Information Theory
Another key progenitor of cognitive science was Claude Shannon, an
electrical engineer at M.I.T. who is usually credited with devising informa-
tion theory. Already as a graduate student at M.I.T. in the late 1930s,
Shannon had arrived at a seminal insight. He saw that the principles of
logic (in terms of true and false propositions) can be used to describe the
two states (on and off) of electromechanical relay switches. In his master's
thesis, Shannon provided an early suggestion that electrical circuits (of the
kind in a computer) could embody fundamental operations of thought. I
shall describe this work-so crucial for all subsequent work with comput-
ers-further in chapter 6.
During the next ten years, working in part with Warren Weaver,
Shannon went on to develop the key notion of information theory: that
information can be thought of in a way entirely divorced from specific
content or subject matter as simply a single decision between two equally
plausible alternatives. The basic unit of information is the bit (short for
"binary digit"): that is, the amount of information required to select one
message from two equally probable alternatives. Thus, the choice of a
message from among eight equally probable alternatives required three
bits of information: the first bit narrowed the choice from one of eight to
one of four; the second, from one of four to one of two; the third selects
one of the remaining alternatives. Wiener explained the importance of this
way of conceptualization: "Information is information, not matter or en-
ergy. No materialism which does not admit this can survive at the present
day" (Wiener 1961, p. 132).
Thanks to Wiener's insights, it became possible to think of informa-
tion apart from a particular transmission device: one could focus instead
on the efficacy of any communication of messages via any mechanism, and
one could consider cognitive processes apart from any particular embodi-
ment-an opportunity upon which psychologists would soon seize as they
sought to describe the mechanisms underlying the processing of any kind
21
I I THE CoGNITIVE REvoLUTION
of information. Only very recently have cognitive scientists begun to

wonder whether they can, in fact, afford to treat all information equiva-
lently and to ignore issues of content.
Neuropsychological Syndromes
A comparable contribution to an incipient cognitive science came
from a remote and unexpected scientific comer-the profiles of cognitive
incapacities following damage to the human brain. Paradoxically, this area
of science relies heavily on the travesties of war. As in the era of the First
World War, much was learned during the Second World War about
aphasia (language deficit), agnosia (difficulty in recognition), and other
forms of mental pathology consequent upon injury to the brain. Laborato-
ries in New York, Oxford, Paris, Berlin, and Moscow were all busily
engaged in working with victims of brain damage. When the neuropsycho-
logical researchers began to communicate their findings to one another,
considerable convergence was noted even across cultural and linguistic
boundaries. For instance, aphasia assumed similar forms despite wide dif-
ferences across languages. There was, it seemed, much more regularity in
the organization of cognitive capacities in the nervous system than was
allowed for by wholly environmental accounts of mental processes. Fur-
thermore, the patterns of breakdown could not be readily explained in
terms of simple stimulus-response disruption. Rather, in many cases, the
hierarchy of behavioral responses was altered. For example, in certain
forms of aphasia, the general sentence frame was preserved, but subjects
could not correctly slot individual words into the frame. In other aphasias,
the sentence frame broke down, but individual content words carried
meaning. Thus was struck yet another blow against reflex-arc models of
thought. At the same time, the particular profiles of abilities and disabili-
ties that emerge in the wake of brain damage provided many pregnant
suggestions about how the human mind might be organized in normal
individuals.
By the late 1940s, in areas as diverse as communication engineering
and neuropsychology, certain themes were emerging principally in the
United States, Great Britain, and the Soviet Union. Though I have stressed
the American version of this story, comparable accounts could be pre-
sented from other national perspectives as well. Scholars in these fields
were not only writing but were eagerly meeting with one another to
discuss the many exciting new perspectives. Herbert Simon, ultimately one
of the founders of cognitive science but then a graduate student at the
University of Chicago, recalls a kind of "invisible college" in the 1940s
(Simon 1982). He knew McCulloch at Chicago; he knew of Shannon's
master's thesis at M.I.T.; he knew that Wiener and von Neumann were
22
working on issues in symbolic logic which had grown out of the philo-
sophical writings of Whitehead, Russell, and Frege. Simon himself was
studying at Chicago with Rudolf Camap, who was then putting forth key
notions about the syntax of logic. Such leading biologists (and Hixon
symposiasts) as Ralph Gerard, Heinrich Kluver, Roger Sperry, and Paul
Weiss were working in nearby laboratories on issues of the nervous sys-
tem. Many of the same influences were rubbing off during this period on
Jerome Bruner, Noam Chomsky, John McCarthy, George Miller, Allen
Newell, and other founders of cognitive science.
Catalytic Encounters and Influential Writings
By the 1940s, then, the principal intellectual capital on which cognitive

science was to be constructed had already emerged. A few scholars like
Norbert Wiener attempted a tentative intellectual synthesis, and more
than a few-ranging from students like Herbert Simon to masters like John
von Neumann-sensed the imminent emergence of a new field (or fields)
of study. There was still the resistance implicit in the behaviorist credo, as
well as some doubts that the human mind would be able to study itself
as effectively as it had studied matter and genetics; but these factors did
not suffice to dampen the enthusiasm of those who sensed the vastness of
the prize awaiting the Newton of human cognition.
The intellectual history of this era reveals many meetings among those
interested in matters of cognition as well as a significant number of publi-
cations that helped to promote a new interdisciplinary science of the mind.
It is possible, of course, that cognitive science could have come into being
-and perhaps even have assumed its present form-in the absence of
these conferences, books, and articles. But particularly when scholars seek
to join forces across often remote disciplines, it is crucial for them to have
the opportunity to get together regularly, to question one another, and to
discover those aspects of scientific method, prejudice, and hunch that are
often invisible in the written record.
The Hixon Symposium, then, was but one of many conferences held
among cognitively oriented scientists during the 1940s and 1950s. To be
sure, it was especially important for our story because of two factors: its
linking of the brain and the computer and its relentless challenging of the
then-prevalent behaviorism. Nonetheless, in any history of this new field,
it is necessary to cite a few other circumstances under which aspiring
cognitive scientists met one another.
In the scientific annals of this period, the name of the Josiah P. Macy
23
Foundation looms large. In the winter of 1944, John von Neumann and
Norbert Wiener convened a meeting at Princeton of all those interested in
what later came to be called "cybernetics." Present at the Macy-sponsored
event were many of the scholars already introduced in this narrative.
Wiener later recalled, "At the end of the meeting it had become clear to
all that there was a substantial common basis of ideas between the workers
in the different fields, that people in each group could already use notions
which had been better developed by the others, and that some attempt
should be made to achieve a common vocabulary" (1961, p. 15).
Building on these initial contacts, Warren McCulloch arranged with
the Macy Foundation in the spring of 1946 for a series of meetings on the
problems of feedback. "The idea has been to get together a group of
modest size, not exceeding some twenty in number, of workers in various
related fields and to hold them together for two successive days in aU-day
series of informal papers, discussions, and meals together, until they had
had the opportunity to thresh out their differences and to make progress
in thinking along the same lines" (Wiener 1961, p. 18). Ultimately there
were ten such meetings, about one a year, of what was originally the
Conference for Circular Causal and Feedback Mechanisms in Biological
and Social Systems-soon (and happily) shortened, at Wiener's urging, to
the Conference on Cybernetics. In the transcripts of these conferences, one
discerns ample evidence of scholars informing one another as well as first
intimations of interesting and sometimes unexpected projects. For exam-
ple, it was in discussions at the Macy meetings that the anthropologist
Gregory Bateson first encountered ideas about feedback which he was to
mine in his "double-bind" theory of schizophrenia.
Activity was especially intense in the Boston and Princeton areas and
in California. During the early 1950s, J. Robert Oppenheimer, director of
the Princeton Institute for Advanced Study (of which von Neumann was
a permanent member) became interested in the application of some of
these new ideas in the field of psychology. He regularly invited a group
of psychologists to visit the institute and report on recent developments
in their field. Among those who spent a year there were George Miller and
Jerome Bruner, gifted young psychologists who would shortly play a
fundamental role in the launching of cognitive science.
Again, there was difficult-to-anticipate but promising cross-fertiliza-
tion of ideas. Oppenheimer was particularly interested in analogies be-
tween the problems of perception, as they are viewed by the psychologist,
and issues of observation, which had come to loom large in atomic and
subatomic physics, once one began to work at the atomic and the sub-
atomic levels. He had been pondering the disturbing implications of the
indeterminacy principle, according to which it is impossible to ascertain the
24
position and the velocity of a particle without affecting it during the course
of measurement. Meanwhile, Bruner had been studying the effects of an
observer's attitude and expectations on putatively 0bjective data." One
11
11
day Oppenheimer remarked to him, Perception as you psychologists
study it can't, after all, be different from observation in physics, can it?"
(quoted in Bruner 1983, pp. 95-96).
In Boston, discussion of these cognitive themes was continuing at
M.I. T ., and at the associated Lincoln Laboratories, where a group of young
engineers and psychologists had assembled to work on applied problems,
such as early warning signals in the case of bomb attacks. At nearby
Harvard in the prestigious Society of Fellows, the influence of behaviorist
thinking was dominant among the senior fellows, but the young junior
fellows, including the linguist Noam Chomsky and the mathematician
Marvin Minsky, were already proceeding in different (and anti-behavio-
rist) theoretical directions (Miller 1982). The Ford Foundation, having
decided to help stimulate work in the behavioral sciences, established a
Center for Advanced Study in the Behavioral Sciences in Palo Alto and
also provided funding for a significant proportion (perhaps one third) of
all the research psychologists in America. At the Rand Corporation in
Southern California, groups of mathematicians and engineers were work-
ing on the development of computing machines. Two young scientists,
Allen Newell and Herbert Simon, had begun to talk about the possibilities
of creating machines that could genuinely think. And, again, there was a
British version as well-the Ratio Club, which commenced in 1949. Cen-
tral to the Ratio Club was the notion of processing information in animals
and machines. Members included physiologists, engineers, physicians, and
psychologists with interests in the mind or "minding." Turing occasionally
attended meetings. The group (which met for several years) had the in-
triguing rule that any member who reached the rank of full professor must
be expelled, because he would then have potential control over other
members (McCorduck 1979, p. 78).
In addition to these many face-to-face encounters among those con-

cerned with cognitive matters, there appeared in the late 1940s and early
1950s several books from different quarters which helped to bring the
emerging interdisciplinary ideas to wider attention. One such book, per-
haps the closest analogy in writing to the Hixon Symposium, was W. Ross
Ashby's Design for a Brain (1952).
Ashby, a British physician and mathematician, wished to account for
human mental activity in a mechanistic manner. He sought to show how,
using only logical axiomatic methods, one could design a machine capable
of adaptive behavior or learning. In the proper behaviorist fashion of the
25
day, Ashby deliberately avoided talking of anything like consciousness or

purposeful behavior. Instead, he directed his attention to the way in which
an organism can effect a transition from chaos to stability, thereby enhanc-
ing the possibility of survival. Stability can come about because "the
machine is a self-organizing system, a system that responds to stimuli,
changing its behavior and sometimes its shape in order to achieve stabil-
ity" (McCorduck 1979, p. 83). Ashby's work intrigued young scholars-
like George Miller, Marvin Minsky, Allen Newell, and Herbert Simon-
for he was not interested merely in making a machine that worked well.
"My aim," Ashby declared, "is simply to copy the living brain. In particu-
lar if the living brain fails in certain characteristic ways, then 1 want my
artificial brain to fail too. 1 have attempted to deduce what is necessary,
what properties the nervous system must have if it is to behave at once
mechanistically and adaptively" (1952, pp. v, 130). It was the scope of
Ashby's aspirations, the doggedly logical way in which he proceeded, and
his refusal to "finesse" possible differences between human and mechani-
cal behavior which caught the attention of aspiring cognitive scientists.
Indeed, Ashby's maddening adherence to the strictest behaviorist and
mechanistic canons served as an additional spur to younger investigators:
his challenge continues to hang, at least spiritually, above the desks of
many of today's cognitive scientists.
From more remote quarters began to appear books relevant to the
discussions in the emerging cognitive sciences. For instance, in the area of
linguistics, Roman Jakobson and his colleagues published their first
findings about the distinctive features of language-the units or building
blocks out of which the phonemes (or basic sounds) of language are con-
structed (Jakobson and Halle 1956). In neuropsychology, Donald Hebb
described the developing nervous system so as to account for many aspects
of visual perception and also to illuminate processes of learning and the
growth and subsequent decline of intelligence (Hebb 1949). In anthropol-
ogy, Gregory Bateson introduced his notions about feedback systems in
social systems-for example, among members of a family (Bateson et al.
1956). New mathematical innovations, such as Markov processes and sto-
chastic models, quickly came to the attention of young workers in the
social sciences. And a few names which had garnered attention on the
Continent began to command increasing respect in the Anglo-American
community-Frederic Bartlett, Claude Levi-Strauss, Alexander Luria, Jean
Piaget, Lev Vygotsky.
But all this is by way of stagesetting. The basic ideas for cognitive
science were immanent in the early papers of McCulloch, Turing, von
Neumann, Wiener, and Pitts and were being heatedly debated at the Macy
conferences, the Ratio Club, Harvard's Society of Fellows, and various
26
other institutions and venues. Important papers and books were being
written and discussed. Still, all of this activity was going on, in a sense,
outside established fields of study. It was extracurricular and considered
a bit odd by those in the mainstream-behaviorist psychology, structural
linguistics, functionalist social anthropology, the neuropsychology of ani-
mal learning. It would take more dramatic events to shake these fields to
their foundation-events that were not long in coming.
27
3
Cognitive Science:
The First Decades
A Consensual Birthdate
Seldom have amateur historians achieved such consensus. There has been
nearly unanimous agreement among the surviving principals that cognitive
science was officially recognized around 1956. The psychologist George A.
Miller (1979) has even fixed the date, 11 September 1956.
Why this date? Miller focuses on the Symposium on Information
Theory held at the Massachusetts Institute of Technology on 10-12 Sep-
tember 1956 and attended by many leading figures in the communication
and the human sciences. The second day stands out in Miller's mind
because of two featured papers. The first, presented by Allen Newell and
Herbert Simon, described the "Logic Theory Machine," the first complete
proof of a theorem ever carried out on a computing machine. The second
paper, by the young linguist Noam Chomsky, outlined "Three Models of
Language." Chomsky showed that a model of language production derived
from Oaude Shannon's information-theoretical approach could not possi-
bly be applied successfully to "natural language," and went on to exhibit
his own approach to grammar, based on linguistic transformations. As
Miller recalls, "Other linguists had said language has all the formal preci-
sions of mathematics, but Chomsky was the first linguist to make good on
the claim. I think that was what excited all of us" (1979, p. 8). Not inciden-
tally, that day George Miller also delivered a seminal paper, outlining his
claim that the capacity of human short-term memory is limited to approxi-
mately seven entries. Miller summed up his reactions:
28
Cognitive Science: The First Decades
I went away from the Symposium with a strong conviction, more intuitive
than rationat that human experimental psychology, theoretical linguistics, and
computer simulation of cognitive processes were all pieces of a larger whole, and
that the future would see progressive elaboration and coordination of their shared
concerns .... I have been working toward a cognitive science for about twenty years
beginning before I knew what to call it. (1979, p. 9)
Miller's testimony is corroborated by other witnesses. From the ranks

of psychology, Jerome Bruner declares, "New metaphors were coming into
being in those mid-1950s and one of the most compelling was that of
computing. . . . My "Generation" created and nurtured the Cognitive
Revolution-a revolution whose limits we still cannot fathom" (1983, pp.
274, 277). Michael Posner concludes, "This mix of ideas about cognition
was ignited by the information processing language that arrived in psy-
chology in the early 1950s" (Posner and Shulman 1979, p. 374). And
George Mandler suggests:
For reasons-that are obscure at present, the various tensions and inadequacies
of the first half of the twentieth century cooperated to produce a new movement
in psychology that first adopted the label of information processing and after
became known as modern cognitive psychology. And it all happened in the five
year period between 1955 and 1960. Cognitive science started during that five year
period, a happening that is just beginning to become obvious to its practitioners.
(1981, p. 9)
Finally, in their history of the period, computer scientists Allen Newell and
Herbert Simon declare:
Within the last dozen years a general change in scientific outlook has oc-
curred, consonant with the point of view represented here. One can date the change
roughly from 1956: in psychology, by the appearance of Bruner, Goodnow, and
Austin's Study of Thinking and George Miller's "The magical number seven"; in
linguistics, by Noam Chomsky's "Three models of language"; and in computer
science, by our own paper on the Logical Theory Machine. (1972, p. 4)
This impressive congruence stresses a few seminal publications, ema-

nating (not surprisingly perhaps) from the same small group of investiga-
tors. In fact, however, the list of relevant publications is almost endless.
As far as general cognitive scientific publications are concerned, John von
Neumann's posthumous book, The Computer and the Brain (1958), should
head the list. In this book-actually a set of commissioned lectures which
von Neumann became too ill to deliver-the pioneering computer scientist
developed many of the themes originally touched upon in his Hixon Sym-
posium contribution. He included a discussion of various kinds of com put-
29
ers and analyzed the idea of a program, the operation of memory in com-
puters, and the possibility of machines that replicate themselves.
Relevant research emanated from each of the fields that I have desig-
nated as contributing cognitive sciences.* The witnesses I have just quoted
noted the principal texts in the fields of psychology, linguistics, and artifi-
cial intelligence, and many more entries could be added. Neuroscientists
were beginning to record impulses from single neurons in the nervous
system. At M.I.T., Warren McCulloch's research team, led by the neuro-
physiologists Jerome Lettvin and Humberto Maturana, recorded from the
retina of the frog. They were able to show that neurons were responsive
to extremely specific forms of information such as "bug-like" small dark
spots which moved across their receptive fields, three to five degrees in
extent. Also in the late 1950s, a rival team of investigators, David Hubel
and Torsten Wiesel at Harvard, began to record from cells in the visual
cortex of the cat. They located nerve cells that responded to specific infor-
mation, including brightness, contrast, binocularity, and the orientation of
lines. These lines of research, eventually honored in 1981 by a Nobel Prize,
called attention to the extreme specificity encoded in the nervous system.
The mid 1950s were also special in the field of anthropology. At this
time, the first publications by Harold Conklin, Ward Goodenough, and
Floyd Lounsbury appeared in the newly emerging field of cognitive an-
thropology, or ethnosemantics. Researchers undertook systematic collec-
tion of data concerning the naming, classifying, and concept-forming abili-
ties of people living in remote cultures, and then sought to describe in
formal terms the nature of these linguistic and cognitive practices. These
studies documented the great variety of cognitive practices found around
the world, even as they strongly suggested that the relevant cognitive
processes are similar everywhere.
In addition, in the summer of 1956, a group of young scholars, trained
in mathematics and logic and interested in the problem-solving potentials
of computers, gathered at Dartmouth College to discuss their mutual inter-
ests. Present at Dartmouth were most of the scholars working in what
came to be termed "artificial intelligence," including the four men gener-
ally deemed to be its founding fathers: John McCarthy, Marvin Minsky,
Allen Newell, and Herbert Simon. During the summer institute, these
scientists, along with other leading investigators, reviewed ideas for pro-
grams that would solve problems, recognize patterns, play games, and
reason logically, and laid out the principal issues to be discussed in coming
years. Though no synthesis emerged from these discussions, the partici-
pants seem to have set up a permanent kind of "in group" centered at the
•Full bibliographical references to these lines of research will be provided at appropriate
points in the text.
30
Cognitive Science: The Rrst Decades
M.I.T., Stanford, and Carnegie-Mellon campuses. To artificial intelligence,
this session in the summer of 1956 was as central as the meeting at M.I.T.
among communication scientists a few months later.
Scholars removed from empirical science were also pondering the
implications of the new machines. Working at Princeton, the American
philosopher Hilary Putnam (1960) put forth an innovative set of notions.
As he described it, the development of Turing-machine notions and the
invention of the computer helped to solve-or to dissolve-the classical
mind-body problem. It was apparent that different programs, on the same
or on different computers, could carry out structurally identical problem-
solving operations. Thus, the logical operations themselves (the "soft-
ware") could be described quite apart from the particular "hardware" on
which they happened to be implemented. Put more technically, the
"logical description" of a Turing machine includes no specification of its
physical embodiment.
The analogy to the human system and to human thought processes
was clear. The human brain (or "bodily states") corresponded to the com-
putational hardware; patterns of thinking or problem solving ("mental
states") could be described entirely separately from the particular constitu-
tion of the human nervous system. Moreover, human beings, no less than
computers, harbored programs; and the same symbolic language could be
invoked to describe programs in both entities. Such notions not only
clarified the epistemological implications of the various demonstrations in
artificial intelligence; they also brought contemporary philosophy and em-
pirical work in the cognitive sciences into much closer contact.
One other significant line of work, falling outside cognitive science as
usually defined, is the ethological approach to animal behavior which had
evolved in Europe during the 1930s and 1940s thanks to the efforts of
Konrad Lorenz (1935) and Niko Tinbergen (1951). At the very time that
American comparative psychologists were adhering closely to controlled
laboratory settings, European ethologists had concluded that animals
should be studied in their natural habitat. Observing carefully under these
naturalistic conditions, and gradually performing informal experiments on
the spot, the ethologists revealed the extraordinary fit between animals
and their natural environment, the characteristic Umwelf (or world view)
of each species, and the particular stimuli (or releasers) that catalyze dra-
matic developmental milestones during "critical" or "sensitive" periods.
Ethology has remained to some extent a European rather than an American
specialty. Still, the willingness to sample wider swaths of behavior in
naturally occurring settings had a liberating influence on the types of
concept and the modes of exploration that came to be tolerated in cognitive
studies.
31
The 1960s: Picking Up Steam
The seeds planted in the 1950s sprouted swiftly in the next decade. Gov-
ernmental and private sources provided significant financial support. Set-
ting the intellectual tone were the leading researchers who had launched
the key lines of study of the 1950s, as well as a set of gifted students who
were drawn to the cognitive fields, much in the way that physics and
biology had lured the keenest minds of earlier generations. Two principal
figures in this Selling of cognition" were Jerome Bruner and George
11
Miller, who in 1960 founded at Harvard the Center for Cognitive Studies.
The Center, as story has it, began when these two psychologists ap-
proached the dean of the faculty, McGeorge Bundy, and asked him to help
create a research center devoted to the nature of knowledge. Bundy report-
edly responded, "And how does that differ from what Harvard University
does?" (quoted in Bruner 1983, p. 123). Bundy gave his approval, and
Bruner and Miller succeeded in getting funds from the Carnegie Corpora-
tion, whose president at that time, the psychologist John Gardner, was
sympathetic to new initiatives in the behavioral sciences.
Thereafter, for over ten years, the Harvard Center served as a locale
where visiting scholars were invited for a sabbatical, and where graduate
and postdoctorate students flocked in order to sample the newest thinking
in the cognitive areas. A list of visitors to the Center reads like a Who's
Who in Cognitive Science: nearly everyone visited at one time or another,
and many spent a semester or a year in residence. And while the actual
projects and products of the Center were probably not indispensable for
the life of the field, there is hardly a younger person in the field who was
not influenced by the Center's presence, by the ideas that were bandied
about there, and by the way in which they were implemented in subse-
quent research. Indeed, psychologists Michael Posner and Gordon Shul-
man (1979) locate the inception of the cognitive sciences at the Harvard
Center.
During the 1960s, books and other publications made available the
ideas from the Center and from other research sites. George Miller-
together with his colleagues Karl Pribram, a neuroscientist, and Eugene
Galanter, a mathematically oriented psychologist-opened the decade
with a book that had a tremendous impact on psychology and allied fields
-a slim volume entitled Plans and the Structure of Behavior {1960). In it the
authors sounded the death knell for standard behaviorism with its discred-
ited reflex arc and, instead, called for a cybernetic approach to behavior in
terms of actions, feedback loops, and readjustments of action in the light
32
of feedback. To replace the reflex arc, they proposed a unit of activity
called a "TOTE unit" (for "Test-Operate-Test-Exit"): an important prop-
erty of a TOTE unit was that it could itself be embedded within the
hierarchical structure of an encompassing TOTE unit. As a vehicle for
conceptualizing such TOTE units, the authors selected the computer with
its programs. If a computer could have a goal (or a set of goals), a means
for carrying out the goal, a means for verifying that the goal has been
carried out, and then the option of either progressing to a new goal or
terminating behavior, models of human beings deserved no less. The com-
puter made it legitimate in theory to describe human beings in terms of
plans (hierarchically organized processes), images (the total available
knowledge of the world), goals, and other mentalistic conceptions; and by
their ringing endorsement, these three leading scientists now made it legit-
imate in practice to abandon constricted talk of stimulus and response in
favor of more open-ended, interactive, and purposeful models.
The impact of this way of thinking became evident a few years later
when textbooks in cognitive psychology began to appear. By far the most
influential was Cognitive Psychology by the computer-literate experimental
psychologist Ulric Neisser (1967). Neisser put forth a highly "construc-
tive" view of human activity. On his account, all cognition, from the first
moment of perception onward, involves inventive analytic and synthesiz-
ing processes. He paid tribute to computer scientists for countenancing talk
of an "executive" and to information scientists for discussing accession,
processing, and transformation of data. But at the same time, Neisser
resisted uncritical acceptance of the computer-information form of analy-
sis. In his view, objective calculation of how many bits of information can
be processed is not relevant to psychology, because human beings are
selective in their attention as a pure channel such as a telephone cannot
be. Neisser expressed similar skeptical reservations about the claims sur-
rounding computer programs:
None of [these programs] does even remote justice to the complexity of

human mental processes. Unlike men, "artificially intelligent" programs tend to be
single minded, undistractable, and unemotional .... This book can be construed
as an extensive argument against models of this kind, and also against other
simplistic theories of the cognitive processes. (1967, p. 9)
After Neisser, it was possible to buy the cognitive science approach in

general and still join into vigorous controversies with "true believers."
Enthusiasts of the power of simulation were scarcely silent during this
period. In his 1969 Compton lectures, The Sciences of the Artificial, Herbert
Simon provided a philosophical exposition of his approach: as he phrased
33
it, both the computer and the human mind should be thought of as ~~sym
bol systems"-physical entities that process, transform, elaborate, and, in
other ways, manipulate symbols of various sorts. And, in 1972, Allen
Newell and Herbert Simon published their magnum opus, the monumental
Human Problem Solving, in which they described the ugeneral problem solver"
programs, provided an explanation of their approach to cognitive studies,
and included a historical addendum detailing their claims to primacy in
this area of study.
Textbooks and books of readings were appearing in other subfields of
cognitive science as well. An extremely influential collection was Jerry
Fodor and Jerrold Katz's collection, The Structure of Language (1964), which
anthologized articles representing the Chomskian point of view in philoso-
phy, psychology, and linguistics, and attempted to document why this
approach, rather than earlier forays into language, was likely to be the
appropriate scientific stance. In artificial intelligence, Edward Feigenbaum
and Julian Feldman put out a collection called Computers and Thought (1963),
which presented many of the best-running programs of the era; while their
collection had a definite ~~carnegie slant," a rival anthology, Semantic Infor-
mation Processing, edited by Marvin Minsky in 1968, emphasized the M.I.T.
position. And, in the area of cognitive anthropology, in addition to influen-
tial writings by Kimball Romney and Roy D'Andrade (1964), Stephen
Tyler's textbook Cognitive Anthropology made its debut in 1969.
But by 1969, the number of slots in short-term memory had been
exceeded-without the benefit of chunking, one could no longer enumer-
ate the important monographs, papers, and personalities in the cognitive
sciences. (In fact, though my list of citations may seem distressingly long,
I have really only scratched the surface of cognitive science, circa 1970.)
There was tremendous activity in several fields, and a feeling of definite
progress as well. As one enthusiastic participant at a conference declared:
We may be at the start of a major intellectual adventure: somewhere compara-

ble to the position in which physics stood toward the end of the Renaissance, with
lots of discoveries waiting to be made and the beginning of an inkling of an idea
of how to go about making them. It turned out, in the case of the early development
of modern physics that the advancement of the science involved developing new
kinds of intellectual sophistication: new mathematics, a new ontology, and a new
view of scientific method. My guess is that the same sort of evolution is required
in the present case (and, by the way, in much the same time scale). Probably now,
as then, it will be an uphill battle against obsolescent intellectual and institutional
habits. (Sloan Foundation 1976, p. 10)
When the amount of activity in a field has risen to this point, with
an aura of excitement about impending breakthroughs, human beings
34
CogniHve Science: The First Decades
often found some sort of an organization or otherwise mark the new
enterprise. Such was happening in cognitive science in the early and mid-
dle 1970s. The moment was ripe for the coalescing of individuals, interests,
and disciplines into an organizational structure.
The Sloan Initiative
At this time, fate intervened in the guise of a large New York-based private
foundation interested in science-the Alfred P. Sloan Foundation. The
Sloan Foundation funds what it terms "Particular Programs," in which it
invests a sizable amount of money in an area over a few years' time, in the
hope of stimulating significant progress. ln the early 1970s, a Particular
Program had been launched in the neurosciences: a collection of disciplines
that explore the nervous system-ranging from neuropsychology and
neurophysiology to neuroanatomy and neurochemistry. Researchers
drawn from disparate fields were stimulated by such funding to explore
common concepts and common organizational frameworks. Now Sloan
was casting about for an analogous field, preferably in the sciences, in
which to invest a comparable sum.
From conversations with officers of the Sloan Foundation, and from
the pubiished record, it is possible to reconstruct the principal events that
led to the Sloan Foundation's involvement with cognitive science. In early
1975, the foundation was contemplating the support of programs in several
fields; but by late 1975, a Particular Program in the cognitive sciences was
the major one under active consideration. During the following year, meet-
ings were held where major cognitive scientists shared their views. Possi-
bly sensing the imminent infusion of money into the field, nearly every
scientist invited by the Sloan Foundation managed to juggle his or her
schedule to attend the meetings. Though there was certainly criticism
voiced of the new cognitive science movement, most participants (who
were admittedly interested parties) stressed the promise of the field and the
need for flexible research and training support.
While recognizing that cognitive science was not as mature as
neuroscience at the time of the foundation's commitment to the latter
field, officers concluded that "nonetheless, there is every indication,
confirmed by the many authorities involved in primary explorations, that
many areas of the cognitive sciences are converging, and, moreover, there
is a correspondingly important need to develop lines of communication
35
from area to area so that research tools and techniques can be shared in
building a body of theoretical knowledge" (Sloan Foundation 1976, p. 6).
After deliberating, the foundation decided to embark on a five-to-seven-
year program, involving commitments of up to fifteen million dollars.
(This commitment was ultimately increased to twenty million dollars.)
The investment took the form, initially, of small grants to many research
institutions and, ultimately, of a few large-scale grants to major universi-
ties.
Like the spur provided by the Macy Foundation a generation earlier,
the Sloan Foundation's initiative had a catalytic effect on the field. As more
than one person quipped, "Suddenly I woke up and discovered that I had
been a cognitive scientist all of my life." In short order the journal Cognitive
Science was founded-its first issue appearing in January 1977; and soon
thereafter, in 1979, a society of the same name was founded. Donald
Norman of the University of California in San Diego was instrumental in
both endeavors. The society held its first annual meeting, amid great fan-
fare, in La Jolla, California, in August 1979. Programs, courses, newsletters,
and allied scholarly paraphernalia arose around the country and abroad.
There were even books about the cognitive sciences, including a popular
account, The Universe Within, by Morton Hunt (1982) and my own historical
essay, also supported by the Sloan Foundation.
Declaring the birth of a field had a bracing effect on those who discov-
ered that they were in it, either centrally or peripherally, but by no means
ensured any consensus, let alone appreciable scientific progress. Patrons
are almost always necessary, though they do not necessarily suffice, to
found a field or create a consensus. Indeed, tensions about what the field
is, who understands it, who threatens it, and in what direction it ought to
go were encountered at every phase of the Sloan Foundation's involvement
(and have continued to be to this day).
Symptomatic of the controversy engendered by the Sloan Founda-
tion's support of research in cognitive science was the reaction to a report
commissioned by the foundation in 1978. This State of the Art Report
(soon dubbed "SOAP" for short) was drafted by a dozen leading scholars
in the field, with input from another score of advisers. In the view of the
authors, "What has brought the field into existence is a common research
objective: to discover the representational and computational capacities of
the mind and their structural and functional representation in the brain"
(1978, p. 6). The authors prepared a sketch of the interrelations among the
six constituent fields-the cognitive hexagon, as it was labeled. Through
the use of unbroken and broken lines, an effort was made to indicate the
connections between fields which had already been forged, and to suggest
the kinds of connection which could be but had not yet been effected.
36
I
I
I
I
Artificial
Intelligence
Connections among the Cognitive Sciences

KEY: Unbroken lines = strong interdisciplinary ties
Broken lines = weak interdisciplinary ties
In my view, the authors of the SOAP document made a serious effort

to survey principal lines of research and to provide a general charter for
work in cognitive science, setting forth its principal assumptions. Then,
using the example of how individuals from different cultures give names
to colors, these authors illustrated how different disciplines combine their
insights. (I'll flesh out this example of color naming in chapter 12.) How-
ever, the community-at-large adopted a distinctly negative view of the
report. In fact, such virulent opposition was expressed by so many readers
that, counter to original plans, the document was never published. I think
this negative reaction came from the fact that each reader approached the
document from the perspective of his or her own discipline and research
program. In an effort to be reasonably ecumenical, the authors simply
ensured that most readers would find their own work slighted. Moreover,
there is as yet no agreed-upon research paradigm-no consensual set of
assumptions or methods-and so cognitive scientists tend to project their
own favorite paradigms onto the field as a whole. In view of these factors,
it was probably not possible in 1978 to write a document that would have
won the support of a majority of cognitive scientists.
It would be desirable, of course, for a consensus mysteriously to
37
emerge, thanks to the largesse of the Sloan Foundation, or for some latter-
day Newton or Darwin to bring order into the field of cognitive science.
In the absence, however, of either of these miraculous events, it is left to
those of us who wish to understand cognitive science to come up with our
own tentative formulation of the field. In the opening chapter of this book,
I presented a working definition of cognitive science and alluded to five
key components of the field. Now that I have sketched out some of the
intellectual forces that led to the launching of cognitive science some three
decades ago, I want to revisit these themes in somewhat more detail, in
order to consider some of their implications as well as some of their
problematic aspects. I will then conclude this introductory part by describ-
ing the paradox and the challenge standing at the center of contemporary
cognitive science.
Key Features of Cognitive Science
In my own work I have found it useful to distinguish five features or

"symptoms" of cognitive science: the first two of these represent the "core
assumptions" of the field, while the latter three represent methodological
or strategic features. Not only are these ideas common to most "strong
versions" of cognitive science, but they also serve as specific points of
contention for its critics. I shall list each of these characteristics and then
indicate certain lines of criticism put forth by those most antagonistic to
cognitive science. These criticisms (as voiced by their most vocal adher-
ents) will be expanded upon at appropriate points in the book and re-
viewed in my concluding chapter.
Representations
Cognitive science is predicated on the belief that it is legitimate-in
fact, necessary-to posit a separate level of analysis which can be called
the "level of representation." When working at this level, a scientist traffics
in such representational entities as symbols, rules, images-the stuff of
representation which is found between input and output-and in addition,
explores the ways in which these representational entities are joined,
transformed, or contrasted with one another. This level is necessary in
order to explain the variety of human behavior, action, and thought.
In opting for a representational level, the cognitive scientist is claiming
that certain traditional ways of accounting for human thought are inade-
38
quate. The neuroscientist may choose to talk in terms of nerve cells, the
historian or anthropologist in terms of cultural influences, the ordinary
person or the writer of fiction in terms of the experiential or phenomeno-
logical level. While not questioning the utility of these levels for various
purposes, the cognitive scientist rests his discipline on the assumption that,
for scientific purposes, human cognitive activity must be described in
terms of symbols, schemas, images, ideas, and other forms of mental repre-
sentation.
In terms of ordinary language, it seems unremarkable to talk of human
beings as having ideas, as forming images, as manipulating symbols, im-
ages, or languages in the mind. However, there is a huge gap between the
use of such concepts in ordinary language and their elevation to the level
of acceptable scientific constructs. Cautious theorists want to avoid posit-
ing elements or levels of explanation except when absolutely necessary;
and they also want to be able to describe the structure and the mechanisms
employed at a level before "going public" with its existence. While talk
about the structure and mechanisms of the nervous system is relatively
unproblematic-since its constituent units can (at least in principle) be
seen and probed-agreement to talk of structure and processes at the level
of mental representation has proved far more problematic.
Critics of the representational view are generally drawn from behavi-
orist ranks. Wielders of Ockham's razor, they believe that the construct of
mind does more harm than good; that it makes more sense to talk about
neur_ological structures or about overt behaviors, than about ideas, con-
cepts, or rules; and that dwelling on a representational level is unnecessary,
misleading, or incoherent.
Another line of criticism, less extreme but ultimately as crippling,
accepts the need for common-sense talk about plans, intentions, beliefs,
and the like but sees no need for a separate scientific language and level
of analysis concerned with their mental representation: on this point of
view, one should be able to go directly from plans to the nervous system,
because it is there, ultimately, that all plans or intentions must be repre-
sented. Put in a formula, ordinary language plus neurology eliminate the
need for talk of mental representations.
Of course, among scholars who accept the need for a level of represen-
tation, debates still rage. Indeed, contemporary theoretical talk among
"card-carrying" cognitive scientists amounts, in a sense, to a discussion of
the best ways of conceptualizing mental representations. Some investiga-
tors favor the view that there is but a single form of mental representation
(usually, one that features propositions or statements); some believe in at
least two forms of mental representation-one more like a picture (or
image), the other closer to propositions; still others believe that it is possi-
39
ble to posit multiple forms of mental representation and that it is impossi-

ble to determine which is the correct one.
All cognitive scientists accept the truism that mental processes are
ultimately represented in the central nervous system. But there is deep
disagreement about the relevance of brain science to current work on
cognition. Until recently, the majority viewpoint has held that cognitive
science is best pursued apart from detailed knowledge of the nervous
system-both because such knowledge has not yet been forthcoming and
out of a desire to ensure the legitimacy of a separate level of mental
representation. As the cognitive level becomes more secure, and as more
discoveries are made in the brain sciences, this self-styled distancing may
be reduced. Not surprisingly, neuroscientists (as a group) have shown the
least enthusiasm for a representational account, whereas such an account
is an article of faith among most psychologists, linguists, and computer
scientists.
Computers
While not all cognitive scientists make the computer central to their
daily work, nearly all have been strongly influenced by it. The computer
serves, in the first place, as an "existence-proof": if a man-made machine
can be said to reason, have goals, revise its behavior, transform informa-
tion, and the like, human beings certainly deserve to be characterized in
the same way. There is little doubt that the invention of computers in the
1930s and 1940s, and demonstrations of "thinking" in the computer in the
1950s, were powerfully liberating to scholars concerned with explaining
the human mind.
In addition to serving as a model of human thought, the computer also
serves as a valuable tool to cognitive scientific work: most cognitive scien-
tists use it to analyze their data, and an increasing number attempt to
simulate cognitive processes on it. Indeed, artificial intelligence, the science
built around computer simulation, is considered by many the central disci-
pline in cognitive science and the one most likely to crowd out, or render
superfluous, other older fields of study.
In principle, it is possible to be a cognitive scientist without loving the
computer; but in practice, skepticism about computers generally leads to
skepticism about cognitive science. To some critics, computers are just the
latest of a long series of inadequate models of human cognition (remember
the switchboard, the hydraulic pump, or the hologram) and there is no
reason to think that today's "buzz-model" will meet a happier fate. View-
ing active organisms as "information-processing systems" seems a radical
mistake to such critics. Computers are seen by others as mere playthings
40
CogniHve Science: The Rrst Decades
which interfere with, rather than speed up, efforts to understand human
thought. The fact that one can simulate any behavior in numerous ways
may actually impede the search for the correct description of human behav-
ior and thought. The excessive claims made by proponents of artificial
intelligence are often quoted maliciously by those with little faith in man-
made machines and programs.
Involvement with computers, and belief in their relevance as a model
of human thought, is pervasive in cognitive science; but again, there are
differences across disciplines. Intrinsic involvement with computers is a
reliable gauge of the extent of a discipline's involvement with cognitive
science. Computers are central in artificial intelligence, and only a few
disgruntled computer scientists question the utility of the computer as a
model for human cognition. In the fields of linguistics and psychology, one
will encounter some reservations about a computational approach; and yet
most practitioners of these disciplines do not bother to pick a feud with
computerphiles.
When it comes to the remaining cognitive sciences, however, the
relationship to the computer becomes increasingly problematic. Many
anthropologists and many neuroscientists, irrespective of whether they
happen to use computers in their own research, have yet to be convinced
that the computer serves as a viable model of those aspects of cognition
in which they are interested. Many neuroscientists feel that the brain
will provide the answer in its own terms, without the need for an inter-
vening computer model; many anthropologists feel that the key to
human thought lies in historical and cultural forces that lie external to
the human head and are difficult to conceptualize in computational
terms. As for philosophers, their attitudes toward computers range from
unabashed enthusiasm to virulent skepticism-which makes them a par-
ticularly interesting and important set of informants in any examination
of cognitive science.
De-Emphasis on Affect, Context, Culture, and History

Though mainstream cognitive scientists do not necessarily bear any
animus against the affective realm, against the context that surrounds any
action or thought, or against historical or cultural analyses, in practice they
attempt to factor out these elements to the maximum extent possible. So
even do anthropologists when wearing their cognitive science hats. This
may be a question of practicality: if one were to take into account these
individualizing and phenomenalistic elements, cognitive science might be-
come impossible. In an effort to explain everything, one ends up explaining
nothing. And so, at least provisionally, most cognitive scientists attempt
41
to so define and investigate problems that an adequate account can be

given without resorting to these murky concepts.
Critics of cognitivism have responded in two principal ways. Some
critics hold that factors like affect, history, or context will never be explica-
ble by science: they are inherently humanistic or aesthetic dimensions,
destined to fall within the province of other disciplines or practices. Since
these factors are central to human experience, any science that attempts to
exclude them is doomed from the start. Other critics agree that some or
all of these features are of the essence in human experience, but do not feel
that they are insusceptible to scientific explanation. Their quarrel with an
antiseptic cognitive science is that it is wrong to bracket these dimensions
artificially. Instead, cognitive scientists should from the first put their noses
to the grindstone and incorporate such dimensions fully into their models
of thought and behavior.
Belief in Interdisciplinary Studies

While there may eventually be a single cognitive science, all agree that
it remains far off. Investigators drawn from a given discipline place their
faith in productive interactions with practitioners from other disciplines:
in the tradition of the Hixon and Macy symposiasts, they hope that,
working together, they can achieve more powerful insights than have been
obtained from the perspective of a single discipline. As examples, they
point to current work in visual perception and in linguistic processing
which has come to draw quite naturally on evidence from psychology,
neuroscience, and artificial intelligence-so much so that disciplinary lines
are beginning to blur.
Skeptics feel that you cannot make progress by compounding disci-
plines, and that it is more prudent to place each individual disciplinary
house in order. Since it is also unclear which of the relevant disciplines will
ultimately contribute to a cognitive science, and in which way, much
valuable time may be wasted in ill-considered collaborations. From their
vantage point, it is perfectly all right to have individual cognitive sciences
but ill-considered to legislate a single seamless discipline. At most, there
should be cooperation among disciplines-and never total fusion.
Rootedness in Classical Philosophical Problems

As already indicated, I consider classical philosophical problems to be
a key ingredient in contemporary cognitive science and, in fact, find it
difficult to conceive of cognitive science apart from them. The debates of
the Greek philosophers, as well as of their successors in the Enlightenment,
42
stand out in many pages of cognitive scientific writing. I do not mean that
these traditional questions have necessarily been phrased in the best way,
or even that they can be answered, but rather that they serve as a logical
point of departure for investigations in cognitive science.
In my discussions with cognitive scientists, however, I have found this
precept to be contentious. Nor is it predictable which scientists, or which
science, will agree with a philosophically based formulation of the new
field. Some cognitive scientists from each discipline readily assent to the
importance-indeed, the inevitability-of a philosophical grounding;
while others find the whole philosophical enterprise of the past irrelevant
to their concerns or even damaging to the cognitive scientific effort. We
may well be dealing here with personal views about the utility of reading
and debating classical authorities rather than with fundamental methodol-
ogical aspects of cognitive science. But whatever the reason, cognitive
scientists are scarcely of a single mind when it comes to the importance of
the Meno, of Descartes's Cogito, or of Kant's Critique.
Precisely because the role of philosophy is controversial in the cogni-
tive sciences, it is useful to explore the earlier history of philosophy. Only
such a survey can prove that cognitive scientists-whether or not they are
fully aware of it-are engaged in tackling those issues first identified by
philosophers many decades or even many centuries ago. Scientists will
differ on whether these questions were properly formulated, on whether
philosophers made any significant progress in answering them, and on
whether philosophers today have any proper role in a scientific enterprise.
Indeed, even philosophers are divided on these issues. Still, it is worth
reviewing their positions on these issues, for philosophers have, since
classical times, taken as their special province the definition of human
knowledge. Moreover, they have also pondered the nature and scope of the
cognitive-scientific enterprise, and their conclusions merit serious exami-
nation.
In my own view, each of these symptoms or features of cognitive
science were already discernible in the discussions of the 1940s and were
widespread by the middle 1950s. A cognitive-science text will not neces-
sarily exhibit or illustrate each of the symptoms, but few texts will be
devoid of most of them. What legitimizes talk of cognitive science is the
fact that these features were not in evidence a half-century ago; and to the
extent that they once again pass from the scene, the era of cognitive science
will be at an end.
Comments on the ultimate fate of cognitive science are most properly
left to the conclusion of this study; but as a kind of guidepost to succeeding
chapters, it may be useful to anticipate my principal conclusions. In my
view, the initial intoxication with cognitive science was based on a shrewd
43
hunch: that human thought would tum out to resemble in significant

respects the operations of the computer, and particularly the electronic
serial digital computer which was becoming widespread in the middle of
the century. It is still too early to say to what extent human thought
processes are computational in this sense. Still, if I read the signs right, one
of the chief results of the last few decades has been to call into question
the extent to which higher human thought processes-those which we
might consider most distinctively human-can be adequately approached
in terms of this particular computational model.
Which leads to what I have termed the computational paradox. Paradoxi-
cally, the rigorous application of methods and models drawn from the
computational realm has helped scientists to understand the ways in which
human beings are not very much like these prototypical computers. This
is not to say that no cognitive processes are computerlike-indeed, some
very much resemble the computer. Even less is it to contend that cognitive
processes cannot be modeled on a computer (after all, anything that can be
clearly laid out can be so modeled). It is rather to report that the kind of
systematic, logical, rational view of human cognition that pervaded the
early literature of cognitive science does not adequately describe much of
human thought and behavior. Cognitive science can still go on, but the
question arises about whether one ought to remain on the lookout for more
veridical models of human thought.
Even as cognitive science has spawned a paradox, it has also encoun-
tered a challenge. It seems clear from my investigation that mainstream
cognitive science comfortably encompasses the disciplines of cognitive
psychology, artificial intelligence, and large sections of philosophy and
linguistics. But it seems equally clear that other disciplines mark a bound-
ary for cognitive science. Much of neuroscience proceeds at a level of study
where issues of representation and of the computer-as-model ~re not
encountered. On the opposite end of the spectrum, much of anthropology
has become disaffected with methods drawn from cognitive science, and
there is a widespread (and possibly growing) belief that the issues most
central to anthropology are better handled from a historical or a cultural
or even a literary perspective.
And here inheres the challenge to cognitive science. It is important for
cognitive science to establish its own autonomy and to demonstrate ter-
rains in which computational and representational approaches are valid. I
believe that cognitive science has already succeeded in this endeavor,
though the scope of its enterprise may not be so wide as one would have
wished.
If cognitive scientists want to give a complete account of the most
central features of cognition, however, they (or other scientists) will have
44
to discover or construct the bridges connecting their discipline to neighbor-
ing areas of study-and, specifically, to neuroscience at the lower bound,
so to speak, and to cultural studies at the upper. How to do this (or whether
it can be done at all) is far from clear at this point: but unless the cognitive
aspects of language or perception or problem solving can be joined to the
neuroscientific and anthropological aspects, we will be left with a disem-
bodied and incomplete discipline. Put differently, no one challenges the
autonomy of biology, chemistry, and physics; but unless a single narrative
can be woven from the components of atomic, molecular, and organic
knowledge, the full nature of organic and inorganic matter will remain
obscure.
All this risks getting ahead of our story, however. We have seen in the
preceding pages how different factors present early in the century came
together to form the bedrock of a new discipline. Ultimately, I want to take
a close look at some of the best work in the discipline, so that I can properly
evaluate its current status and its future prospects. To achieve this over-
view, however, it is necessary to consider how the very framing of ques-
tions within cognitive science grows out of philosophical writings of the
past. By the same token, it is necessary to understand the particular histo-
ries, methods, and problems that have characterized the component cogni-
tive sciences. Ultimately this philosophical and historical background has
determined in large meas\.rre the nature and scope of current interdiscipli-
nary cognitive-scientific efforts. In part II of this book, I shall take a careful
look at the several disciplines whose existence made possible the idea of
cognitive science and whose practitioners will determine the success of this
enterprise.
45
PART II
THE COGNITIVE
SCIENCES:
A HISTORICAL
PERSPECTIVE
4
Reason, Experience, and
the Status of Philosophy
Philosophy always buries its undertakers.
-ETIENNE GILSON
There is no department of knowledge in which so little

progress has been made as in that of mental philosophy.
The human mind has been studied as if it were indepen-
dent of the body, and, generally speaking, by philosophers
who possessed a comparatively small share of physical
knowledge. No attempt, indeed, has been made to examine
its phenomena by the light of experiment and observation,
or to analyze them in their abnormal phases.... Without
data, without axioms, without definitions, [the science of
mind] proposes problems which it cannot solve .... The
human mind escapes from the cognizance of sense and
reason, and lies, a waste field with a northern exposure,
upon which every passing speculator casts his mental
tares, choking any of the good seed that may have sprung
up towards maturity.
-DAVID BREWSTER (1854)
Descartes's Mind
In this chapter I begin my survey of the cognitive sciences by examining

the history and the current status of philosophy. This choice is appropriate.
Not only is philosophy the oldest of the cognitive sciences, but its epis-
temological branch has supplied the initial agenda-the list of issues and
49
II I THE CoGNITIVE SciENCES : HISTORICAL PERSPECTIVE
topics upon which empirically oriented cognitive scientists are working

today. Indeed, philosophers have wrestled for centuries with such cur-
rently fashionable issues as the nature of mental representation, the extent
to which human thought is merely a mechanical (as opposed to a spiritual)
process, and the relationship between reason and feeling.
Rene Descartes is perhaps the prototypical philosophical antecedent
of cognitive science. Writing at the very start of the modern era (which,
in part, he helped to define and launch), he is in some ways a throwback
to that confident reliance on one's own intuitions, that commitment to the
centrality of mathematical thinking, that belief in innate ideas which we
associate with the Greeks. At the same time, in his curiosity about the
operation of the sensory system, the nature of brain processes, and the
automaton as a possible model of human nature, he is virtually a contem-
porary figure.
When, in 1623, he sequestered himself in a small farmhouse in Ba-
varia, he was embarking on a program of reflection that was to exert
profound effects on subsequent Western thought. Descartes had become
disenchanted with the systems spun by previous thinkers: "[Philosophy]
has been cultivated for many centuries by the best minds that have ever
lived and nevertheless no single thing is to be found in it which is not the
subject of dispute" (quoted in Wilson 1969, p. 111). To cope with this
uncertainty, Descartes evolved a method of systematic doubt. He deter-
mined to call into question any evidence of which he was not certain, and
found that all he was left with then were his own states of consciousness,
his own doubts. Upon his capacity to doubt and, therefore, to think,
Descartes discerned a secure foundation on which to build a new
philosophy.
The centerpiece of Descartes's philosophy was his own mind and, by
extension, the minds of individuals in general. Through examination of the
contents of his own mind, Descartes felt that he could establish the knowl-
edge that was most valid, least subject to challenge. He searched for
thoughts that were clear and distinct, hence indubitable. A privileged
position was reserved for the ideas of arithmetic and geometry, which
seemed least fettered by doubt, so evident that they must be true. More-
over, these ideas had not arrived at his mind from external sources: rather,
it made sense to think of them as generated by the mind itself:
And what I believe to be more important here is that I find in myself an

infinity of ideas of certain things which cannot be assumed to be pure nothingness,
even though they may have perhaps no existence outside of my thought. These
things are not figments of my imagination, even though it is within my power to
think of them or not to think of them; on the contrary, they have their own true
and immutable natures. Thus, for example, when I imagine a triangle, even though
50
Reason, Experience, and the Status of Philosophy
there may perhaps be no such figure anywhere in the world outside of my thought,
nor ever have been, nevertheless the figure cannot help having a certain determi-
nate nature . . . or essence, which is immutable and eternal, which I have not
invented and which does not in any way depend upon my mind. (1951, * p. 61)
Mind, in Descartes's view, is special, central to human existence,

basically reliable. The mind stands apart from and operates independently
of the human body, a totally different sort of entity. The body is best
thought of as an automaton, which can be compared to the machines made
by men. It is divisible into parts, and elements could be removed without
altering anything fundamental. But even if one could design an automaton
as complex as a human body, that automaton can never resemble the
human mind, for the mind is unified and not decomposable. Moreover,
unlike a human mind, a bodily machine could never use speech or other
signs in placing its thoughts before other individuals. An automaton might
parrot information, but "it never happens that it arranges its speech in
various ways, in order to reply appropriately to everything that may be
said in its presence, as even the lowest type of man can do" (quoted in
Wilson 1969, p. 138).
Descartes was aware that the positing of two distinct entities-a
rational mind and a mechanical body-made implausible any explanation
of their interaction. How can an immaterial entity control, interact with,
or react to a mechanical substance? He made various stabs at solving this
problem, none of them (as he knew) totally convincing. But in the process
of trying to explain the interaction of mind and body, Descartes became
in effect a physiologically oriented psychologist: he devised models of how
mental states could exist in a world of sensory experience-models featur-
ing physical objects that had to be perceived and handled. He asked:
What must be the fabric of the nerves and muscles of the human body in order
that the animal spirits therein contained should have the power to move the
members ... what changes are necessary in the brain to cause wakefulness, sleep
and dream; how light, sounds, smells, tastes, and all other qualities pertaining to
external objects are able to imprint on it various ideas by the intervention of the
senses. (Quoted in Wilson 1969, p. 137}
And he even proposed one of the first "information-processing" devices:

Descartes's diagram showed how visual sensations are conveyed, through
the retinas, along nerve filaments, into the brain, with signals from the two
eyes being reinverted and fused into a single image on the pineal gland.
There, at an all-crucial juncture, the mind (or soul) could interact with the
body, yielding a complete representation of external reality. The mind
•when relevant, original publication dates are listed in the References (p. 393).
51
could now consciously perceive the image and play upon the brain, some-
what as a musician plays upon an instrument (Fancher 1979, pp. 133-34).
Like many who have reflected on the sources of knowledge, Descartes
had once thought that all experiences and thoughts arise through the
senses. But through his meditations, he had come to devalue the senses and
to attribute all thought and creativity to the mind. While he had to admit
that there were sources for the experiences of his senses, he minimized
their significance:
I cannot doubt that there is in me a certain passive faculty of perceiving, that

is, of receiving and recognizing the ideas of sensible objects; but it would be
valueless to me and I could in no way use it if there were not also in me or in
something else, another active faculty capable of forming and producing these
ideas. (1951, p. 75)
Even as Plato had placed his faith in a mind that could possess (or remem-
ber) all manner of things, Descartes determined that the mind, an active
reasoning entity, was the ultimate arbiter of truth. And he ultimately
attributed ideas to innate rather than to experiential causes:
no ideas of things, in the shape in which we envisage them by thought, are

presented to us by the senses. So much so that in our ideas there is nothing which
was not innate in the mind, or faculty of thinking, except only these circumstances
which point to experience .... They transmitted something which gave the mind
occasion to form these ideas, by means of an innate faculty, at this time rather than
another. (Quoted in Chomsky 1966, p. 67)
If Descartes could have dispensed with external experience altogether, he

would have been pleased to do so; again, like Plato, he attributed human
error and inconstancy to the vagaries of experience, and our rationality,
understanding, and genuine knowledge to the mind reflecting upon its
own ideas. And in so doing, he hurled the strongest possible challenge to
future empiricist philosophers, whose thinking he would provoke.
In his discussion of ideas and the mind, sensory experience and the
body, the power of language and the centrality of an organizing, doubting
self, Descartes formulated an agenda that would dominate philosophical
discussions and affect experimental science in the decades and centuries
that followed. Furthermore, he proposed the vivid and controversial image
of the mind as a rational instrument which, however, cannot be simulated
by any imaginable machine-an image still debated in cognitive science
today. By creating such images and adopting such stances, Descartes
helped launch a period that turned out to be the richest ever in philosophy.
As Alfred North Whitehead was to note, "A brief, and sufficiently accu-
rate, description of the intellectual life of the European races during the
succeeding two centuries and a quarter up to our own times is that they
52
have been living upon the accumulated capital of ideas provided for them
by the genius of the seventeenth century" (1925, p. 42).
The agenda initially posed by the Greeks, and carried forward during
the seventeenth century by Descartes, came to be fervently debated among
the group of philosophers who earned Whitehead's admiration. The initial
empirical responses by Locke, Hume, and Berkeley, the bold synthesis put
forth by Kant, and the eventual challenge to the Kantian tradition were
principal milestones in the philosophical debates about knowledge. Then,
at the very time when the epistemological tradition associated with these
earlier writers was undergoing the most severe criticism, the computer
came on the scene. Very soon newly discovered philosophical perspectives,
which go by such names as "functionalism," and "intentionality," raised
once again the possibility of a respectable epistemology: a cooperative
scientific effort in which the representational nature of thought could be
countenanced, and where traditional philosophical concerns merged with
the work of cognitively oriented scientists.
Two important themes have recurred in philosophy over the past few
centuries. The first involves the tension between rationalists and empiri-
cists. Those of a rationalist persuasion believe that the mind exhibits
powers of reasoning which it imposes upon the world of sensory experi-
ence; empiricists believe that mental processes either reflect, or are con-
structed on the basis of, external sensory impressions. Both Plato and
Descartes embraced the rationalist pole, while many succeeding empiri-
cists reacted to it. In our own era, behaviorists have clung to empiricism,
while cognitivists are likely to embrace some form of rationalism or a
rationalist-empiricist mix.
A second, discipline-oriented theme concerns the actual status of
philosophy within the world of scholarly disciplines and, in particular, its
relation to science. Once again, Descartes and Plato share a common per-
spective: from their confident stance, philosophical reflection is the pri-
mary pursuit, while the observations of more empirically oriented scholars
are given less credence. (Although science was only beginning to be estab-
lished in Descartes's time, he carried out scientific investigations and surely
thought of himself as a scientific worker.) In succeeding generations, the
findings and laws of science became increasingly visible: indeed, some
issues (for example, the essential nature of matter) were so adequately
resolved by the physical sciences that they dropped outside the purview
of philosophy. Eventually many, if not most, philosophers felt the need
both to keep up with scientific discoveries and to justify their activities in
a scientifically respectable manner. In recent decades, as the dominance of
science continues to grow, the value of philosophical investigations has
again been challenged.
From the perspective of some cognitivists, the rise of empirically
53
oriented investigation renders philosophy unnecessary. With science to

answer the philosophical agenda, what is left for philosophers to do? I shall
argue, however, that this conclusion is wrong. I see at work a dialectic
process, where philosophers propose certain issues, empirical disciplines
arise in an attempt to answer them, and then philosophers cooperate with
empirical scientists in interpreting the results and in proposing new lines
for work. The issues raised by Descartes and his contemporaries ultimately
became the concerns of psychologists, linguists, and neuroscientists a few
centuries later; and indeed, Descartes's reflections on the human as a possi-
ble automaton are central to artificial intelligence today. In the light of
empirical results, philosophers have sometimes fundamentally reconcep-
tualized the issues on which they were working; and these reconceptuali-
zations, in tum, have sustained and directed empirical work and aided in
its interpretations. Rather than being the ultimate arbiters, or the ultimate
victims, of scientific work, philosophers have been (and will continue to
be) important handmaidens in the scientific study of cognition.
Empiricist Responses to Descartes
Locke s Competing Model

Descartes's views were soon challenged by a group of British empiri-
cist philosophers, chief among them John Locke, George Berkeley, and
David Hume. Locke began by questioning whether one could accept any
knowledge on the basis of introspective evidence; he looked instead to
sensory experience as the only reliable source of knowledge. He challenged
the belief in innate ideas as being useless and misleading. "The knowledge
of the existence of any other thing we can have only by sensation, " he declared.
"For the having the idea of anything in our mind, no more proves the
existence of that thing, than the picture of a man evidences his being in
the world, or the visions of a dream make thereby a true history" (quoted
in Hanfling 1972, p. 356). And in a famous passage he issued his ultimate
answer to the origins of knowledge:
Let us then suppose the mind to be, as we say, white Paper, void of all
Characters, without any Ideas: How comes it to be furnished? Whence comes it by
that vast store, which the busy and boundless Fancy of Man has painted on it, with
an almost endless variety? Whence has it all the materials of Reason and Knowl-
edge? To this I answer, in one word, From Experience. (Quoted in Herrnstein and
Boring 1965, p. 584)
54
Reason, Experience, and the Status o/ Philosophy
These words sum up the empiricist worldview.
While (on the empiricist account) the experience of the world begins
with perception, it assuredly does not end there. Locke devoted considera-
ble effort to distinguishing among various external qualities (which he
called "primary" and "secondary," "simple" and "complex"). He stressed
how complex ideas grow out of simple ones and the various ways in which
ideas come to be associated with one another. He described the process by
which words come to stand for ideas and make possible abstract or general
ideas-for example, the notion of the general idea of a triangle (as opposed
to a specific triangle with its unique sides and angles). And, in the end, he
posited a person or self able to appreciate these ideas, "a thinking intelligent
being, that has reason and reflection and can consider itself as itself, the
same thinking thing in different time and places" (Copleston 1964, p. 111 ).
Locke's epistemology focused on the opposite end of the universe
from Descartes's-on the experience of objects in the external world. Yet,
Locke was able in the end to construct an organism capable of abstraction
and generalization, one deriving knowledge of a certain kind from the
interplay of these ideas (rather than from a simple comparison with experi-
ence) and culminating in a rational, conscious self. It is not fanciful to
suggest that the Lockean individual bears a strong family resemblance to
the Cartesian individual. But they differ profoundly in the methods by
which each has been achieved.
Berkeley s and Humes Skeptical Tones

Locke's successors voiced skepticism about certain aspects of his
model of man. Berkeley (1929) became so solipsistic that he ultimately
denied the existence of the material world, apart from one's perception of
it. He ridiculed the notion of a "general idea" and challenged the privileged
status of mathematics. He saw language as a barrier to communication, as
fostering belief in entities that do not exist, as causing imprecision in
thought, and as arousing passions rather than reflection. He placed his
faith in the primacy of the experiencing self, the perceiving mind, which
alone makes it possible to have sensations or to conceive of ideas.
Hume's skepticism proceeded in a different but equally devastating
direction. His particular bugbear was the idea of causality as a privileged
link between two events that regularly succeed one another. Hume
demonstrated persuasively that there is no license for inferring from par-
ticular correlations of past events the generalization that one event causes
another. The most that one can assume is that nature will perform the same
in the future as in the past: this assumption involves a customary or
habitual way of thinking on our part, not a necessary, given link. Hume
55
II I THE CoGNITIVE SciENCES : HISTORICAL PERSPECTivE
thus undermined confidence in a rational order, in scientific explanations
of causal appearances.
He was equally brutal toward the Cartesian belief in the centrality of
the mind, questioning the existence of a substance or locale called "the
mind," in which perceptions inhere. To Hume, we know the mind only as
we know matter-by perception. In fact, the mind is best thought of as an
abstract name for a series of ideas-as a kind of theater where several
perceptions successively make their appearance: there is no observing or
controlling soul behind the processes of thought (Wolff 1967, p. 92). As
the historian Will Durant once quipped (1926), Hume destroyed the mind
as quickly as Berkeley destroyed matter.
Despite the rampant skeptical strain among these latter British empiri-
cists, they produced a set of interlocking issues-the nature of sensory
experience, the classification of objects, the role of language, the status of
the individual conscious self-which budding philosophers were expected
to ponder (issues also that are presented to today's fledgling cognitive
scientists).
Before long, while individual positions continued to change, the belief
that these were the central issues was assumed by all participants in the
debate. In fact, the belief that our conceptual apparatus is built up from
qualities perceived in the external world became so entrenched that for
many years researchers, no less than laypersons, had difficulty in coming
up with rival accounts of why we perceive the world in the way we do.
But, with all that, the nature of mind continued to dominate discussion.
Even Hume, the most skeptical of all, adhered to this central agenda:
It becomes, therefore, no inconsiderable part of science ... to know the

different operations of the mind, to separate them from each other, to class them
under their proper heads, and to correct all that seeming disorder in which they
lie involved when made the object of reflection and inquiry.... It cannot be
doubted that the mind is endowed with several powers and faculties, that these
powers are distinct from one another, and that what is really distinct to the
immediate perception may be distinguished by reflection and, consequently, that
there is a truth and falsehood in all propositions on this subject, and a truth and
falsehood which lie not beyond the compass of human understanding. (1955, p. 22)
Kant and Foundational Philosophy
By the late eighteenth century, the German scholar Immanuel Kant was
faced with rival alternatives: one, favored by the British empiricists,
viewed thought as merely an instrument to reflect or build upon mundane
56
experience; the other, favored by Descartes, Leibnitz, and others in the
Continental tradition, stressed the universal province of thought as the
organizer and revealer of all possibilities. Indeed, the German Gottfried
Leibnitz had directly answered the empiricists: addressing Locke's state-
ment that there is nothing in the mind that has not been in the senses,
Leibnitz added the telling Cartesian rejoinder, "Nothing but the mind
[intellect] itself." The empiricists were suspicious of a priori statements and
proofs (which they dismissed as unrevealing tautologies), while the ration-
alists searched for universal principles embodied in pure thought.
In his monumental Critique of Pure Reason, first published in 1781, Kant
strove to synthesize these rationalist and empiricist points of view. To
accomplish this task, he had to confront the question whether there might
exist knowledge that is necessarily so-hence, a priori- but is also in some
sense dependent upon experience, not just conjured up t~utologically in
the mind. He had to apply the systematic analysis of the Leibnitz and
Cartesian tradition to the vagaries of daily experience, which had seemed
regnant to Locke, Hume, and Berkeley. Kant chose to examine both ends
of this polarity: to understand the nature of experience and, even more
important, to dissect the nature of mind.
In what was probably the crucial step in this process, Kant had to
understand what permits the mind to apprehend experience in the way it
does, and to yield necessary knowledge. In analyzing what he called the
synthetic a priori, Kant had to show how knowledge begins with experience
(thus not being purely analytic) and yet does not arise out of, or come from,
it (thus not being purely a posteriori). He had to explain the sources of
arithmetic, geometry, Newtonian physics, traditional logic-those ulti-
mate achievements of human minds, which seemed beyond dispute-that
is, necessary-once they had been discovered.
A point of departure for Kant was the individual ego-the individual
with his own awareness and judgment. In this sense, Kant was a Cartesian:
he began with the knowledge of the ego-the transcendental self. "I think"
must be capable of accompanying all propositions. Nor was this transcend-
ing self a passive instrument: Kant, more than did his predecessors, saw
the mind as an active organ of understanding which molds and coordinates
sensations and ideas, transforming the chaotic multiplicity of experience
into the ordered unity of thought. The transcendental self is always the
active subject: unknown in itself, following rules built into its own opera-
tion, it makes our experience possible.
Nonetheless, this self, this organizing mental entity, does not operate
in autistic fashion. It depends upon and is stimulated by the outside world.
The closest we can get to truth are the assertions we can make about
information that arises under optimal conditions through our sensible
natures. On these assumptions Kant developed a framework detailing how
57
knowledge proceeds from and builds out of the "raw" materials of sensory
experience.
Kant's model can be parsed into several parts. Initially, there is the
concrete sensory world-the sensory manifold-which exists external to
the subject and from which one's knowledge must begin. But this sensory
world cannot be perceived directly: there is no privileged access to the
thing itself (das Ding an sich). We must deal always with phenomena-
appearances-and not with noumena-the unknowable external world.
Any phenomenon consists of sensations, which are caused by particular
objects themselves: the form of the phenomenon is due to our subjective
apparatus which orders the manifold in certain ways. The actual forms of
intuition which we use as human beings in apprehending the world are our
"spectacles." For instance, we must see things in terms of their embedded-
ness in space and in time: we have no choice.
But over and above the immanent properties of space and time, our
understanding brings to bear a set of what Kant (following Aristotle) called
"categories of thought." These elementary concepts of the pure under-
standing-such as quantify (unity, plurality, and totality); qualify (reality,
negation, and limitation); relation (substance and accident, cause-and-
effect, and reciprocity); modality (possibility, existence, and necessity)-
constitute the mental equipment, the pure synthesizing concepts with
which human understanding is endowed. These alone allow the individual
to make sense of his experiences.
These categories seem remote from bright red patches or fresh red
cherries-examples that have dominated empiricists' accounts of sensory
experience. Kant may have thought so as well, because he devised yet
another level of analysis-that of the schemas, or schemata-interposed be-
tween the raw sensory information and the abstract a priori categories. As
he put it, with characteristic obscurity, "This representation of a universal
procedure of the imagination in providing an image for a concept, I entitle
the schema of this concept" (quoted in Wolff 1967, p. 76). In devising this
explanatory apparatus, Kant sought to determine the circumstances under
which the categories can find concrete employment. A schema serves as a
mediating representation which is intellectual in one sense, sensible in
another. Thus, a schema is directly activated in terms of sensory experience
and yet can be plausibly thought to provide an interpretation of that
experience. As a cognitive scientist might put it today, Kant had entered
the world of "mental representation."
While the nature and the operation of the schemas pose difficulties
even for Kantian scholars, epistemologists need such a level of analysis to
deal with particular instances while mobilizing the abstract categories. The
schemas are in part rules and thereby are linked to pure understanding; but
58
they are also in part images and so are linked to empirical perception. The
schema of each category determines the condition by which it is applicable
to the objects of experience in general. Thus, the schema for the category
of quantity is number; the schema for quality is degree of intensity; the
schema for relation is permanence in time; and so on. Moreover, the
schemas are also found at the level of concrete experience. Because we can
apply the concept of dog to Fido, we must be able to produce in our
imagination the schematic representation of a dog-the schema of the
concept here being distinguishable from the concept itself. Schematic the-
ory thus demonstrates how the categories can have empirical conse-
quences.
While Kant did not publish empirical results of any kind, and his
writing on these topics is notorious for its difficulty, his thinking has left
its mark on most theoretical writings in cognitive science today. What
Kant was groping for (in my view) was a way of describing the level of
representation, for a terminology that might account for the way in which
knowledge must be represented in any entity so as to relate to the physical
world, on the one hand, and to the world of the inborn mental architecture,
on the other. And while we may use less tortuous language nowadays for
addressing these issues, I think that no one has gone beyond Kant in
sketching the nature of the problem and in proposing a plausible solution,
on the conceptual level, for how knowledge is possible-or, as he would
have put it, for the conditions for the possibility of knowledge.
Kant had no illusions about the enormousness of the task-but he had
a good deal of confidence about his capacity to negotiate it. He declared
in the preface to the first edition of the Critique of Pure Reason, "I venture to
assert that there is not a single metaphysical problem which has not been
solved, or for the solution of which the key at least has not been supplied"
(quoted in Russell 1961, p. 680). In the preface to the second edition, he
compared himself to Copernicus and opined that, by placing knowledge
in the mind of the active thinker, he had effected a Copernican revolution
in philosophy.
Kant's belief in his own powers eventually proved contagious, and
many philosophically oriented scholars came to believe that he had deli-
mited the domain within which the acquisition of knowledge is possible.
Kant's charting of territory included three reasons why a science of psy-
chology was not possible: the fact that the mind is affected while studying
itself; the lack of spatial extent essential for any experimentation; and the
absence of a mathematical basis, necessary for all science. Kant concluded
that "psychology can, therefore, never become anything more than a his-
torical (and, as such, as much as possible) systematic natural doctrine of
the internal sense, i.e., a natural description of the soul, but not a science
59
II I THE CoGNITIVE SciENCEs : HISTORICAL PERSPECTIVE
of the soul, nor even a psychological experimental doctrine" (quoted in

Watson 1979, p. 88). For many years this warning inhibited researchers
interested in cognitive matters.
The Logical-Empiricist Program
Kant's program, his conclusions, and even his warnings exerted an unpar-
alleled effect on ensuing research. Whereas each of the earlier epistemolo-
gists had soon aroused a formidable opponent, Kant's epistemology re-
mained without major competition for years. He had apparently reconciled
the two competing strands in philosophical writing-the primacy of
thought and the primacy of experience; and his argument was sufficiently
difficult, his critiques of standard rationalist and empiricist positions suffi-
ciently powerful, his examination of the nature of mentation sufficiently
revolutionary, his view of philosophy sufficiently reassuring, that it was
many years before the weaknesses in his formulation became apparent.
Thanks to Kant, philosophy for a time occupied a special place in the
firmament of scholarly disciplines. In light of his supposed demonstration
of what knowledge was, of how its attainment was possible, and of why
Newtonian science was necessary, Kant conferred upon philosophy the
status of a "super" or "meta" discipline. Every scholar was expected to
study this foundational subject, which dealt with the most fundamental
aspects of knowledge; and no educated person would admit to not having
read Kant. Anyone who entertained grave doubts about the utility of
philosophical analysis usually kept silent. And it continued to be assumed
that science and philosophy could march onward together.
It was not the philosophers who adopted Kant's speculative mode, but
rather logicians, who eventually challenged Kantian epistemology. Kant
had declared that, since Aristotle, logic "has not been able to advance a
single step and is to all appearances a closed and completed body of
doctrine" (quoted in Abel1976, p. 52). In fact, however, in the century and
a quarter following Kant, thinkers like George Boole, Gottlob Frege, Gi-
useppe Peano, Charles Sanders Peirce, and, ultimately Bertrand Russell
and Alfred North Whitehead had effected many reformulations in logic.
Indeed, rather than being a set of approximate procedures, logic became
a non-empirical science, into which new findings could be regularly ab-
sorbed. There were equally dramatic advances in mathematics and physics,
as scholars like Niels Bohr, George Cantor, Albert Einstein, and Max
Planck rewrote our understanding both of the world of mathematics and
60
of the physical universe. Eventually, progress in the fields that Kant
held in highest regard called into question many of his fundamental as-
sumptions: the given nature of time and space, the source of mathematical
propositions, the unmodifiable rules of logic, the impossibility of
psychology.
The work that had the greatest effect on the mainstream of epistemol-
ogy was carried out in Cambridge, England, at the beginning of the twen-
tieth century by Alfred North Whitehead and Bertrand Russell. These
mathematically oriented logicians had been impressed by the work of the
German logician Gottlob Frege who had undertaken to show that arithme-
tic could be reduced to logic. As Frege had written, "Even an inference like
that from n to n + 1, which on the face of it is peculiar to mathematics,
is based on the general laws of logic" (quoted in Hanfling 1972, p. 103).
Building on Frege's insights, Whitehead and Russell sought to derive all
of mathematics from the basic laws of logic, as these laws had been refor-
mulated in the century since Kant's time. Indeed, this program was in part
explicitly anti-Kantian, for the philosophers wanted to discredit the syn-
thetic a priori notion of mathematical knowledge as being dependent upon
experience. With respect to this part of their program, they were largely
successful.
But Whitehead and Russell's overall program exerted its most revolu-
tionary effect on philosophy. Their work led to much closer ties among
empirical science, logic, and mathematics, so that the boundaries between
these pursuits could no longer be sharply delineated. Whereas Kant saw
philosophy as the ultimate foundational discipline, the standard by which
other sciences had to be judged, Whitehead and Russell saw all of these
disciplinary forays as part of the same larger enterprise. Indeed, Russell
believed that many-if not most-traditional philosophical questions
could be expressed in logical terms and either solved in these terms, or
shown to be insoluble. As he confidently declared:
Modern analytical empiricism ... differs from that of Locke, Berkeley, and
Hume by its incorporation of mathematics and its development of a powerful
logical technique. It is thus able, in regard to certain problems, to achieve definite
answers, which have the quality of science rather than of philosophy. It has the
advantage, as compared with the philosophies of the system-builders, of being able
to tackle its problems one at a time, instead of having to invent at one stroke a block
theory of the whole universe. Its methods, in this respect, resemble those of
science. I have no doubt that, in so far as philosophical knowledge is possible, it
is by such methods that it must be sought: I have also no doubt that, by these
methods, many ancient problems are completely soluble .... Take such questions
as: What is number? What are space and time? What is mind, and what is matter?
I do not say that we can here and now give definitive answers to all these ancient
questions, but I do say that a method has been discovered by which, as in science,
61
we can make successive approximations to the truth, in which each new stage
results from an improvement, not a rejection, of what has gone before. (Russell
1961, pp. 788, 789)
As contributors to the scientific enterprise, Whitehead and Russell began

with an acceptance of the world of sensory experience, which any individ-
ual initially has to confront. They sought to close the gap between what
is immediately known and what can be known through inference by using
the tools of logic: it should be possible eventually to give an account of
the external world by means of a logical construction from sensory data.
In their own way, the two Englishmen had put forth a program as ambi-
tious as the ones propounded by those "system-builders" whom Russell
had challenged. At the same time, his expressed belief that his approach
could ultimately provide empirically valid answers to long-standing philo-
sophical questions-questions dating back to the time of Socrates-puts
him within the ranks of cognitive science.
Just as the epistemological agenda of the last century had been dic-
tated by Kant's monumental treatise, much of the philosophical program
of this century-particularly in the Anglo-American world with which I
am most concerned-can be seen as an effort to pursue Russell and White-
head's program. As we shall see, their blend of scientific philosophy has
been found seriously deficient. Nonetheless, many observers feel that
much was gained by the effort to approach human sensory experience with
the methods of logical empiricism. Even more important for present pur-
poses, the logical-empiricist school strongly influenced many pioneering
cognitive scientists.
One of the first thinkers inspired by the Whitehead-Russell program
was the young Austrian philosopher Ludwig Wittgenstein. In his Trac-
tafus (1961), which he sent to Russell for a first reading in 1918, Wittgen-
stein attempted to demonstrate a logical structure implicit in language. As
he saw it, language provides, in a sense, a picture of the structure of facts
-indeed, a picture of the world. More specifically, the propositions of
language are the perceptual expression of thoughts, and thoughts are logi-
cal pictures of facts. Wittgenstein thus posited a formal correspondence
between configurations of objects in the world, thoughts in the mind, and
words in language.
It has been said of necessity that Aristotle thought of it as arising from
things; that to Kant it derived from the structure of our minds; while the
Wittgenstein of the Tractafus saw it as arising from language. Wittgenstein
injected language into Russell and Whitehead's program. In Wittgenstein's
program, philosophy becomes the activity of clarifying propositions,
though the propositions clarified by philosophical analysis are not them-
selves propositions of philosophy but rather are nonphilosophical proposi-
62
tions about the world (Kenny 1973). At the same time that it clarifies the
propositions of natural science, philosophy can also expose as nonsense
metaphysical questions about existence.
As is well known, Wittgenstein himself came first to doubt and then
to renounce virtually completely his "picture theory" of language and the
world (Wittgenstein 1968). But his program of analyzing language logi-
cally and trying to relate that analysis to the perceived world became a
pivotal ingredient of the program of philosophy adopted by the "Vienna
circle" of logical empiricists in the period between the world wars. Scholars
like Herbert Feigl, Otto Neurath, Morris Schlick, and especially Rudolf
Camap tried in earnest to tie together what A.]. Ayer called "language,
truth, and logic" (1936). The general goal of the circle was to see which
traditional philosophical questions could be rephrased in formal terms and
which could not be. Those that could not be (for instance, issues about
celestial angels) were branded as metaphysical and banned forthwith from
further discussion. Those propositions that lent themselves to treatment in
logical terms (for instance, ones about geometric angles) would then be
examined to see whether they could be verified and thus added to our
storehouse of what is true.
One major ingredient of the program of the logical empiricists was
verificafionism, a doctrine that assumed that empirical (nonlogical) state-
ments can be verified under ideal conditions of inquiry. All that is needed
is a way of measuring and verifying what is being talked about. As Carnap
put it, one must observe the circumstance under which people use a propo-
sition in order to determine the truth of that proposition. Indeed, the
meaning of a proposition is its method of verification. Another doctrine of
the program was physicalism. The Vienna circle believed that propositions
ordinarily construed as referring to mental states turn out to be logically
equivalent to those referring to overt behavior. Indeed, every sentence of
psychology could be reformulated as a description of the physical behavior
of humans or other animals. Finally, the Vienna circle rejected the notion
of philosophy as a special or foundational area of study: as far as these
philosophers were concerned, there was just one empirical science, of
which philosophy was part (Ayer 1982, p. 128); its role was to describe and
criticize the language of science.
Armed with this set of ideas, Rudolf Carnap determined to put into
practice the logical-empiricist beliefs. Russell had talked about defining
the world from experience, through logical construction; and Carnap
sought to translate, into the language of sensory data, all sentences pertain-
ing to the world. For every sentence about material objects, a correspond-
ing sentence about sensory data can express the basic phenomenal ele-
ments of the experience of the object.
Having devised sensory-centered sentences of this sort, the next step
63
in the Camap program was to examine the relationship among sentences.

In Camap's view, traditional philosophical problems had best be consid-
ered by examining the theoretical relations between whole systems of
sentences: the nature of an entity is determined once one has ascertained
the inferential relationships between sentences using that term and other
sentences in that language. Ultimately, Camap believed that much of
philosophy could be reduced to such issues of logical syntax: when errors
in syntax are avoided (through the use of logical analysis), a philosophical
problem can either be solved or be shown to be insoluble.
As I have already suggested, Camap's program (and that of the Vienna
circle) did not work out at all; yet the attempt has been considered signifi-
cant. Thus, Nelson Goodman, an influential American philosopher who
has criticized the Vienna program, says of Camap's magnum opus:
The Aufbau [the logical construction of the world) brings to philosophy the
powerful techniques of modern logic, along with unprecedented standards of ex-
plicitness, coherence, and rigor. It applies to [philosophy] the new methods and
principles [which had been brought to bear] upon mathematics .... The potential
importance to philosophy is comparable to the importance of the Euclidean de-
ductive method into geometry.... [It] is still one of the fullest examples we
have of the logical treatment of problems in nonmathematical philosophy. (1972,
p. 22)
Thus, with Russell and Camap, Goodman believes that the techniques
from logical analysis have the potential to solve many-even if not all-
existing philosophical problems, and that attacking the problems one at a
time is the best means for achieving success.
There is another reason the saga of logical empiricism is highly rele-
vant to a history of cognitive science. Inspired by the Whitehead-Russell
tradition, Camap and his colleagues sought to express everyday scientific
findings in terms of the basic elements of logic. Loosely speaking, we might
see this effort as a ferreting out of the logical structure of science and, more
especially, of the language used in science-as a study of the syntax of
science. As I see it, a major ingredient in ongoing work in the cognitive
sciences has been cast in the image of logical empiricism: that is, the vision
of a syntax-a set of symbols and the rules for their concatenation-that
might underline the operations of the mind (and a correlative discomfort
with issues of mental content). Thus, when Noam Chomsky (1965) posits
the basic operations of a grammar, when Richard Montague (Thomason
1974) examines the logic of semantics, when Allen Newell and Herbert
Simon (1972) simulate human reasoning on a computer, or when Jerome
Bruner (1973) and George Miller (1956) seek to decipher the rules of
classification, or "chunking," they are trying to decipher a logic-perhaps
64
the logic-of the mind. This vision comes through even more clearly in the
writings of Jerry Fodor, who explicitly searches for a "language of
thought" and even appropriates certain of Camap's methods. Thus, a
model that proved inadequate for the scientific enterprise as a whole still
motivates research in circumscribed cognitive domains.
In applying logical methods to the world of empirical experience, the
members of the Vienna circle combined the spirit of rationalism and em-
piricism. They also sought to establish a place for philosophy within the
world of laboratory science by providing the tools for analyzing scientific
statements and practices. In hindsight, their procedures were too artifi-
cially constrained, and their idea of theory too tied to behaviorist and
positivist strictures. And yet even those cognitivists who see themselves
as doctrinarily opposed to the letter of the Vienna program often partake
of the spirit of their approach.
The Unraveling of Logical Empiricism and the Revised

Roles of Philosophy
During the late 1930s and early 1940s, a group of philosophers met regu-
larly at Harvard. Among them were the Harvard philosophers Nelson
Goodman and W. V. 0. Quine, as well as such frequent visitors as Rudolf
Camap, Bertrand Russell, and Alfred Tarski, all three from war-tom
Europe. These philosophers had all been involved with logical empiricism
but now subjected nearly all of its assumptions to increasingly severe
criticism. First of all, they questioned whether it is possible to talk about
"raw information": that is, pure sensory data which can be inspected and
built upon. Knowledge came to be seen increasingly as a matter of using
propositions, and it was now dubious practice to talk about knowing the
pure, the immediate, the given. Even more tellingly, the whole logical
apparatus devised to analyze "meaning" and "truth" came to be chal-
lenged. While the distinction between analytic and synthetic truths had
been cherished since the time of Kant, Quine {1953) showed that it is
ultimately untenable: the notion of meaning itself is not clear enough to
justify the attribution to certain statements of "true by virtue of meaning
alone" or "true by definition." Moreover, the logical (or analytic) compo-
nents of a scientific theory cannot be sufficiently disentangled from the
empirical components to allow them to be regarded as subject to different
criteria of truth. After all, Euclid's principles would certainly seem to have
been a priori truths; and yet their truth was ultimately undermined by
65
non-Euclidean geometries. Indeed, once one is confronted by others who

do not share one's intuitions, there is no way to establish which truths are
due to language, which to common experience. Ultimately, there are as
many ways to relate language to the world as there are languages, as many
logical syntaxes as there are language systems.
As a result of these and other criticisms, the major American philoso-
pher Hilary Putnam concluded:
Not a single one of the great positive theses of Logical Empiricism (that
Meaning is Method of Verification; that metaphysical propositions are literally
without sense; that Mathematics is True by Convention) has turned out to be
correct. It detracts from the excitement of the fact that, by turning philosophical
theses into linguistic ones [as Carnap had tried to do] ... one can make philosophy
more scientific and settle the truth value of philosophical propositions by hard
scientific research, if the results one obtains are uniformly negative. (1975a, p. 20)
Still, he softened this verdict, in ways revealing for this history, by declar-
ing, "Even if [logical empiricism] failed, modern symbolic logic, a good
deal of modern language theory, and part of contemporary cognitive sci-
ence were all off-shoots of these attempts" (1984, p. 274).
Critics like Quine and Putnam regretted the collapse of logical empiri-
cism and have been sympathetic toward cognitive science, perhaps sensing
that this movement shares at least some of the methods and aspirations of
earlier philosophical work. But by the end of the 1940s, far more strident
voices were being raised about the tradition that had begun with Descartes
and was, after many twists and turns, still discernible in the logical-empiri-
cist camp. The trio of names most vividly associated with this criticism are
Gilbert Ryle, Ludwig Wittgenstein (in his later writings), and]. L. Austin.
Gilbert Ryle
In his justly celebrated The Concept of Mind (1949), Ryle explicitly con-
fronted the "official doctrine" of mentalism that had begun with Descartes.
As Ryle described this doctrine, it entailed the belief that everyone has a
mind and a body; minds are not in space and their operations are not
subject to mechanical laws; the workings of the mind are private and are
accessible only to the person alone; there are, in fact, two different kinds
of existence or status-whatever happens may have a physical or a mental
existence. "Mental happenings occur in insulated fields, known as 'minds'
and there is ... no direct causal connection between what happens in one
mind and what happens in another" (1949, p. 13). Ryle dubbed this doc-
trine, with its many associated assumptions and linguistic turns, as the
"dogma of the Ghost in the Machine." He declared, "I hope to prove that
66
it is entirely false, and false not in detail but in principle" (p. 16), and
devoted a 330-page book to justifying that hope.
Ryle's basic claim was that talk of mind involves a category mistake.
It is perfectly all right to speak about "minds"; but we must not fall into
the trap of thinking that there is a place called "the mind," with its own
locations, events, and so on, any more than we assume that there is a place
called "the university" apart and separate from buildings, roads, lawns,
persons, and other physically specifiable entities. To talk of mind as if it
has a separate existence, was, to Ryle, mistakenly to treat an entity of one
sort (an abstract characterization of a set of dispositions) as if it were itself
another of those dispositions. "The theoretically interesting category-mis-
takes," Ryle pointed out, "are those made by people who are perfectly
competent to apply concepts, at least in the situations in which they are
familiar, but are still liable in their abstract thinking to allocate those
concepts to logical types to which they do not belong" {1949, p. 17). And
the mind becomes such a theoretically interesting mistake because Des-
cartes posited a substance, parallel to body but apart from it, that controls
and is the scene of our mental life.
Having exposed the flaw in the Cartesian position, Ryle proceeded to
show how one can talk, in ways that do not involve category mistakes or
violate the actual state of affairs, about those entities and experiences that
are generally termed "mental." In general, he took a behaviorist stance:
when we talk about a person as having "traits" or inner volitional capaci-
ties, we are simply indicating that people are disposed to behave in certain
ways and are likely to do so in the presence of the appropriate eliciting
circumstances. Ryle questioned whether there are actually happenings in
the mind to which each person has privileged access. Instead, he insisted,
what we can find out about ourselves is in principle no different from the
sorts of things we can find out from other people through observation and
through questioning. To speak of a person's mind is simply to talk of
certain ways in which the incidents of one's life are ordered.
Ryle declared that he was not interested in the issues of how one sees
or understands something if those seeings or understandings involve the
positing of some internal understanding or perceptual mechanisms. The
most that we can do as philosophical analysts is to try to understand the
circumstances under which an individual would report having seen or un-
derstood something; and these circumstances should be accessible across
the full range of reporting individuals. Ryle objected to answers invoking
internal mechanisms ("I understand something because I process certain
information in certain ways") to conceptual questions ("What are the
circumstances under which an individual is likely to report having under-
stood something?"). In his view, the positing of internal mechanisms (a
67
II I THE CoGNITIVE SciENCEs : HisToRICAL PERSPECTIVE
Ia Descartes or Kant) adds nothing to our understanding. Ryle would have

seen little reason to embrace a "representational level"; nor would he have
been sympathetic with contemporary efforts to found a whole science
upon "internal entities" like schemas, rules, or representations whose very
positing he found problematic.
Ludwig Wiffgensfein
Wittgenstein saw that many philosophical conundrums might be
solved by careful attention to the ways in which people use words. In his
earlier writings, he had treated language as a means of understanding the
world-as a privileged means of looking through to the structure of the
world. Now, however, he saw language itself as the spawner of prob-
lematic issues, and the exercise of coming to understand how we use
language as the therapy for philosophical problems. In his later work,
Wittgenstein did not try to solve problems but tried rather to show that
they arise from a network of terms that have evolved in such a way as
to make their disentanglement extremely difficult. As he once com-
mented, his aim in philosophy was to show the fly the way out of the
fly-bottle.
Wittgenstein came to view language as an inherently public or com-
munal activity. One is introduced into language by others in the commu-
nity, and thus one comes to know how to use words. One certainly has
a private experience, for example, of pain, but the use of the word
pain comes from the ways in which, and the circumstances under which,
it is regularly employed by other people. We do not first use the word
pain by naming something we feel; rather, the word is embedded in various
activities that people carry out and in various sorts of things others say in
connection with being hurt.
More generally, Wittgenstein believed, it is instructive to think of
language as a set of games: we proceed from the fact that we are always
involved in many language games-interactions with other individuals in
which we move around sets of linguistic counters; and, like a set of games,
each of these little encounters has its own set of rules. But it is not easy
to ferret out these rules because they overlap with one another: the lan-
guage games constantly mesh. To add to this tangled state of affairs, words
do not have clear and unambiguous meanings. The word game itself has a
family of meanings, with no definition ever sufficient to account for all,
and only all, games. Given the numerous language games occurring at any
one time, and their inherently overlapping nature, it is no wonder that
Wittgenstein despaired of ever solving philosophical problems in the rig-
orous way that he and his Viennese peers had once hoped. It made more
68
sense to try to dissolve the problems altogether, by showing that they had
been deceptively phrased.
Wittgenstein' s attitude toward traditional problems can be gleaned
from his comments about psychology, particularly the version that he had
encountered during his early studies in Vienna. As he saw it, psychologists
were trying to solve problems they did not understand: rather than being
genuinely scientific, these problems were in fact embedded in certain uses
of language. Wittgenstein illustrated the complex and overlapping manner
in which many mentalistic words-believe, imagine, desire-are actually used.
Instead of trying to explain how each of these putative mental operations
actually "works," it would make more sense for psychologists (a Ia Ryle)
to study the relations among such ways of talking about behavior and
experience. In a pessimistic assessment, Wittgenstein asserted:
The confusion and barrenness of psychology is not to be explained by calling

it a young science-the existence of experimental methods makes us think we have
the means of solving the problems that trouble us: though problems and methods
pass one another by .... Not to explain but to accept the psychological phenomena
-that is what is so difficult. (Quoted in Hacking 1982, p. 43)
Wittgenstein felt that investigators interested in questions of psychology

ought to ponder the phenomena in which they are interested and, most
especially, the use of terms in these areas, rather than trying to set up
experiments to answer putatively crucial questions. He drew a sharp anal-
ogy: "People who are constantly asking 'why' are like tourists who stand
in front of a building reading Baedeker [a guide book] and are so busy
reading the history of its construction that they are prevented from see-
ing the building. The tendency to explain instead of merely describing [is
what yields] bad philosophy" (p. 43).
f L. Austin
As if one were needed-a final nail was sunk into the coffin of logical
empiricist philosophy by J. L. Austin, another British philosopher inter-
ested in language. Austin demonstrated convincingly that one cannot sim-
ply accept a sentence at face value, as the logical empiricists wanted to do.
Many, if not most, sentences need to be thought of not only in terms of their
literal meaning (ullerance meaning} but also with respect to the use to which
they are put by the deliverer of the utterance (speaker meaning}. As Austin
explained in his William James lectures, How to Do Things with Words:
it has come to be commonly held that many utterances which look like
statements are either not intended at all, or only intended in part, to record or
69
impart straightforward information about the facts .... Many traditional philo-
sophical perplexities have arisen through a mistake-the mistake of taking as
straightforward statements of fact utterances which are eilhn (in interesting non-
grammatical ways) nonsensical or else intended as something quite different.
(1962, pp. 2-3)
Thus, "It is very hot here" is as likely to be a request to open a window,

or a comment on a tense discussion, as a statement concerning the temper-
ature in a room. Once it had been established that any statement might
have quite different effects depending upon who uttered it, in what con-
text, and for what reason, the notion of a neutral evaluation of sentences
had to be abandoned. Isaiah Berlin recalls hearing Austin present his ideas
at Oxford in the early 1950s:
Freddy [A. J.] Ayer was there, still very keen on the Vienna circle. We all knew
by the time the evening was half-done that Camap and Vienna were finished.
Austin made the distinction between analytic and synthetic propositions a special
case of his more general way of classifying propositions by their illocutionary"
force. It was stunning. (Quoted in Bruner 1982, p. 41)
In urging a focus on the ordinary uses of language, Austin, Ryle, and

Wittgenstein were cautioning against the notion of "philosophy as-a-
super-discipline" which could legislate topics like knowledge, truth, and
science. Perhaps some of these questions might be approached in the
manner of empirical science, but certainly philosophers had no privileged
means for attacking or solving them. If philosophy had any special mission
(and these analysts were skeptical), it was to call attention to the habits
of language that often dominate human thinking and to help unravel some
obscure modes of discussion (including those exhibited by some scien-
tists).
Could anything of the program of traditional philosophy be salvaged?
W. V. 0. Quine, himself one of the master architects of the initial critique,
thinks so. While pointing out that the dream of epistemology as a kind of
"first philosophy" has failed, he insists that there remains a legitimate role
for this activity:
But I think that at this point it may be more useful to say rather that episte-
mology still goes on, though in a new setting and a clarified status. Epistemology,
or something like it, simply falls into place as a chapter of .psychology and hence
of natural science. It studies a natural phenomenon, viz, a physical human subject.
This human subject is accorded a certain experimentally controlled input-certain
patterns of irradiation in assorted frequencies, for instance-and in the fullness of
• Rlocufion refers to the ends for which a statement is uttered-for example, to give an
answer, to make an announcement, or to make a request.
70
Reaspn, Experience, and the Status of Philosophy
time the subject delivers as output a description of the three-dimensional external
world and its history. The relation between the meager input and the torrential
output is a relation that we are prompted to study for somewhat the same reasons
that always prompted epistemology; namely, in order to see how evidence relates
to theory, and in what ways one's theory of nature transcends any available
evidence. (Quoted in Royce and Rozeboom 1972, p. 18)
Quine is proposing a substitute. Oassically, a philosopher considered

how he-or any person--could have a theory, and an understanding of the
world, based on the evidence picked up via his senses. Accumulated cri-
tiques of the last fifty years had shown that there was no way in which
this endeavor could provide an independent justification of our knowledge
of the world. There is no pure sense experience, no clear-cut meaning, no
unambiguous use of language, no privileged syntax, no "prior" philosoph-
ical problem. However, the field of empirical psychology does offer a way
of approaching these issues. As Quine sees it, we no longer dream of
deducing science from sense data: the scientist (whether philosopher or
psychologist) now conducts research in which experimental subjects be-
come the preferred route to discovering how any individual makes sense
of his experiences. Locke and Kant are not completely irrelevant but are
supplanted by experimental investigators of particular regulated interac-
tions between a subject and the world. Quine quotes a metaphor of which
he is fond: that is, the construction and reconstruction of science places
each of us in the position of the mariner who must rebuild his boat, plank
by plank, while staying afloat on it. On this view, cognitive scientists can
be seen as rebuilding the traditional structure of epistemological inquiry.
Richard Rorty: Is Epistemology Necessary?
Surveying the same scene as Quine, Richard Rorty reaches a far less san-
guine conclusion. In his much-discussed Philosophy and the Mirror of Na-
ture (1979}, Rorty questions the whole enterprise of epistemology. He
suggests a radically different and far more modest view of philosophy, one
that might even undermine the programs pursued by Wittgenstein, Ryle,
and Austin, and certainly one much less ambitious and far less optimistic
than Quine's. Rorty's impressive critique of all of philosophy since the
Greek times not only has intrinsic interest but also raises profound ques-
tions about the enterprise called cognitive science, at least as I have charac-
terized it here.
According to Rorty, philosophers have thought of their discipline as
71
confronting perennial questions, enigmas that arise as soon as one begins

to reflect. Among these familiar chestnuts are: How does a person come to
know anything? And, what are the limitations of knowledge? To delve
into these foundational issues is to discover something about the mind:
(Philosophy] understands the foundations of knowledge and it finds these

foundations in a study of man-as-knower, of the "mental processes" or the" activ-
ity of representation" which make knowledge possible. To know is to represent
accurately what is outside the mind; so to understand the possibility and nature
of knowledge is to understand the way in which the mind is able to construct such
representations. Philosophy's central concern is to be a general theory of represen-
tation .... We owe the notion of a "theory of knowledge" based on an understand-
ing of "mental processes" to the seventeenth century, and especially to Locke. We
owe the notion of "the mind" as a separate entity in which "processes" occur to
the same period, and especially to Descartes. We owe the notion of philosophy as
a tribunal of pure reason, upholding or denying the claims of the rest of culture,
to the eighteenth century and especially to Kant, but this Kantian notion presup-
posed general assent to Lockean notions of mental processes and Cartesian notions
of mental substance. (1979, pp. 3-4)
Rorty considers this habit of analysis-this invocation of certain images

for thinking about mind-as pernicious an endeavor as does Ryle. In fact,
elsewhere he goes further:
I want to suggest that the concept of mind is the blur with which Western
intellectuals became obsessed when they finally gave up on the blur which was the
theologian's concept of God. The ineffability of the mental serves the same cultural
function as the ineffability of the divine-it vaguely suggests that science does not
have the last word. (Rorty 1982b, p. 31)
In Rorty's view, this set of beliefs is the culmination of a long history

of thinking in misguided ways about certain kinds of activity. It is all right
to label these activities misguided-as Ryle has done: but the only way to
exorcise this "ghost" completely is to review its history, much as a thera-
pist helps a patient relive the history of past activities and blunders. Thus,
a major part of Rorty's effort involves a reconstruction, or deconstruction,
of Western philosophical thinking in an effort to show the various ways
in which Philosophy Went Wrong.
The developmental sequence expounded by Rorty can be summarized
as follows: In Greek times, beliefs were determined through face-to-face
confrontation with the object of belief. Thus the slave boy in the
Meno discovered the truths of geometry through examination of a triangle.
Knowledge was a matter of having accurate representations of ideal forms
that could not be directly observed.
The next stage, courtesy of Descartes, held that our knowledge de-
72
pends upon the activity of a quasi-visual faculty-which Rorty dubs the
Mirror of Nature-and consists of an assembly of accurate representations.
These representations are in the mind; and an "inner eye" surveys them
hoping to find some mark that will testify to their fidelity. Though the
empiricists differed with Descartes in many respects, they preserved his
mentalistic notion of the mind as a separate region which inspects ideas.
Locke introduced a fatal confusion when he confounded an explanation of
how information gets into our consciousness-a classical psychological
question-with a justification of why we believe what we believe-a classi-
cal philosophical question.
Unlike his predecessnrs, Immanuel Kant understood the impossibility
of anyone's having direct access to things: that is, one has knowledge of
propositions about objects not about objects themselves. But, in an effort
to locate the most accurate representations, Kant ultimately posited a
special privileged set of representations that cannot be doubted. According
to Rorty's revisionist account, these representations came to be considered
the foundation of all knowledge; and thus Kant granted to philosophers
the pre-eminent position for making statements about the world and for
regulating inquiry.
In Rorty's view, in the years since Kant's time much of philosophy has
sought to retain this vision. But the vision has undergone severe jolts,
thanks to a series of compelling critiques of traditional epistemology.
There was Wittgenstein's attack on the centrality and legitimacy of classi-
cal philosophical problems. There was the pragmatist John Dewey, who
insisted that one should try to use knowledge in a practical way rather than
to strive after the chimera of objective knowledge. There was the
phenomenologist Martin Heidegger, who dissected the various images and
metaphors that have obsessed Western philosophy since Greek times. And
perhaps most tellingly, there were the failures of both the Russell-White-
head program and the Vienna circle to arrive at secure knowledge through
a logical construction from sense data.
Rorty's own interpretation of recent philosophical history relies espe-
cially heavily on criticisms to which I have already alluded. He arrives at
the following conclusion: There is no way to account for the validity of
our beliefs by examining the relation between ideas and their objects:
rather, justification is a social process, an extended conversation, whereby
we try to convince others of what we believe. We understand the nature
of knowledge when we understand that knowledge amounts to justifica-
tion of our belief, and not to an increasingly accurate representation of
reality. As Rorty concludes, "If assertions are justified by society rather
than by the character of the inner representation they express, there is no
point in attempting to isolate privileged representations" (1979, p. 174).
73
II I THE CoGNITIVE SciENCEs : HisTORICAL PERSPECTIVE
Perhaps because he did not fully appreciate the radical implications

of his own (and others') demonstrations, Quine continued to believe that
epistemology can endure through the good offices of the psychologist
studying his "knowing subject." But in Rorty's opinion, psychology can
in no way succeed in doing what epistemology has failed to do. In an
intriguing chapter, Rorty attempts to save psychology from the excessive
claims made by thinkers like Quine who see it as a next best attempt to
answer the very philosophical questions that have turned out to be ill
conceived and outmoded. Rorty asserts that he has no objection to the idea
of ideas per se-what he calls the "idea idea." As the constructs of scientists,
ideas in the mind are no more or less reputable than neurons in the brain.
The damage done by the "idea idea" was done by the pseudo explanation
of epistemological authority: by the claim that the eyes of the mind have
direct acquaintance with special entities like meanings and sense data
(1979, p. 209).
Rorty chides critics like Wittgenstein and his disciple Norman Mal-
colm for being too severe on psychology. Indeed, Rorty does not object to
the psychologist Peter Dodwell's (1971) contention that psychologists are
in the best position to decide which issues to investigate, and to investigate
them in the manner, and with the concepts, that make most sense to them
-so long, that is, as psychologists do not suggest that these investigations
provide answers to questions like, "How is abstraction, or recognition, or
constancy possible?" or, "Where in the nervous system does information
begin to be processed?"
But it would be a severe misreading of Rorty, I submit, to conclude
that psychology receives a clean bill of health. In another section of his
book, Rorty imagines the Antipodeans emanating from a society where
one talks not about ideas or feelings or beliefs but just about the stimula-
tion of various fibers in the brain. Moreover, the Antipodeans cannot
understand why we earthlings insist on such mentalistic talk: for them, the
only thing that gives rise to experience is the stimulation of parts of the
brain. Rorty seems entirely in sympathy with this materialist position,
where psychological experience is equated with neurology. Consider these
various remarks in his book:
Psychologists should be more mechanistic rather than less .... They should
cut straight through the mental to the neurophysiological. (1979, p. 217)
This is to say that if physiology were simpler and more obvious than it is,
nobody would have felt the need for psychology. (p. 237)
We can imagine machines in which it would be easier to find out what the
machine was up to by opening it up and looking than by reading the program.
(p. 238)
74
If we have psychophysiology to cover causal mechanisms, and the sociology
and history of science to note the occasions on which observation sentences are
invoked or dodged in constructing and dismantling theories, then epistemology has
nothing to do. We would think that this result would be congenial to Quine, but
in fact he resists it. (p. 225)
And, in the context of discussing the effect that his Antipodeans might
have on philosophy, Rorty adds, "The disappearance of psychology as a
discipline distinct from neurology, and similar cultural developments,
might eventually free us from the image of the Mirror of Nature much
more effectively than philosophers' identity theories" (p. 121}.
To summarize, Rorty is unable or unmotivated to come up with any
arguments in principle against psychology, but feels that the discipline
might well never have been invented, and that it may well at some time
disappear, to the regret of few. Neurophysiology seems a much more
secure science on which to base one's hope. The issue of an interdiscipli-
nary cognitive science is not addressed directly by Rorty, but some of his
thoughts on this matter can be gleaned from this telltale aside:
Only the assumption, that one day the various taxonomies put together by,
for example, Chomsky, Piaget, Levi-Strauss, Marx, and Freud will all flow together
and spell out one great Universal Language of Nature-an assumption sometimes
attributed to structuralism-would suggest that cognitive psychology had epis-
temological import. But that suggestion would still be as misguided as the sugges-
tion that, since we may predict everything by knowing enough about matter in
motion, a completed neurophysiology will help us demonstrate Galileo's superior-
ity to his contemporaries. The gap between explaining ourselves and justifying
ourselves is just as great whether a programming language or a hardware language
is used in the explanations. (1979, p. 249)
Rorty believes that the particular course followed by Western philos-

ophy is a matter of history, not of necessity. The history of this field-or
even its existence-would have been entirely different if, for example, the
Greeks had not modeled knowledge upon vision; if Descartes had not
effectively invented the mind by putting feelings and beliefs into a single
organ; or if Kant had not set up a tribunal of knowledge and placed his
own philosophical synthesis upon the judge's bench.
Rorty proposes a rather sensational prescription for philosophy. He
believes that epistemology has served its purposes-mostly evil rather
than good-and should now be gently retired from the circle of disciplines.
Moreover, philosophy as a whole should radically reduce its claims to
specialness. Rorty rejects the need for ambitious philosophers like Kant or
Russell who would systematize all knowledge, and plumps instead for
"edifiers" like Wittgenstein and Dewey, who are content to react and to
75
II I THE CoGNITIVE SciENCEs : HisTORICAL PERSPECTivE
interpret. Rorty looks to a more humanistic philosophy in which one

investigates the ideas of philosophers in order better to understand the
nature and limitations of favored ideas. He wants to abandon the belief
(dating back to Kant, if not to Plato) that the philosopher is someone who
knows something special or especially well, or that there is a special philo-
sophical method, technique, or point of view. He seeks to halt philosophi-
cal debates-for instance, between empiricists and rationalists-as based
on faulty premises and of no consequence for either the humanities or the
sciences. In his renegotiated disciplinary terrain, physics can explain the
structure of the external world, neurophysiology can delineate the pro-
cesses whereby we experience ideas and feelings, and sociology and
history will account for the shape of our current beliefs.
And what of cognitive science? So long as it steers clear of philosophi-
cal questions that permit no answer, this new field can presumably con-
tinue its quest for the reasons that we humans experience and process the
world in the way we do. Rorty seems skeptical that answers of scientific
consequence will come from such efforts, at least as they are now being
pursued.
Preserving Philosophy's Purview
In the academic year 1946--47, Karl Popper, a sometime member of the

Vienna circle, was invited by the secretary of the Moral Sciences Club at
Cambridge to read a paper about "some philosophical puzzles." As Popper
recalls, "It was of course clear that this was Wittgenstein's formulation and
that behind it was Wittgenstein's philosophical thesis that there are no
genuine problems in philosophy, only linguistic puzzles. Since this thesis
was among my pet aversions, I decided to speak on 'Are there Philosophi-
cal Problems?'" (1974, p. 122).
Popper began his lecture by expressing his surprise at the initial invi-
tation since, he pointed out, by implicitly denying that philosophical prob-
lems exist, the inviter had taken sides in what was actually a genuine
philosophical problem. This teasing remark aroused Wittgenstein, who
immediately jumped up and exclaimed loudly and angrily, "The Secretary
did exactly as he was told to do. He acted on my own instruction." Popper
ignored the interruption and went oni but repeatedly through the talk,
Wittgenstein "jumped up again, interrupting me, and spoke at length
about puzzles and the nonexistence of philosophical problems" (1974,
p. 123). The exchange became increasingly heated until eventually Witt-
genstein left the room in anger or disgust.
76
Reflecting on this incident some twenty years later, Popper repeated
his claim that there are indeed philosophical problems, and that he may
have even solved some of them. Yet he went on to add, "Nothing seems
less wanted than a simple solution to an age-old philosophical problem.
The view of many philosophers, and, especially, it seems, of Wittgen-
steinians, is that if a problem is soluble, it cannot have been philosophical"
(1974, p. 124).
This confrontation between Popper and Wittgenstein helps to convey
the atmosphere in philosophical circles at the middle of this century. In the
background stood the great figures of the past-Popper's heroes-who
believed that there are genuine philosophical problems, and that if one
ponders them systematically, one should eventually solve them, or at least
make progress toward solving them. Rorty also recognized this group but
believed that these "systematizers" (as he termed them) have been em-
barked on a hopeless and wrong-headed task. Instead, Rorty reserved his
praise for individuals whom he labels "edifiers," who are deeply skeptical
about the legitimacy of such questions and would rather assume a soft,
even teasing approach toward the whole philosophical enterprise. Among
scholars in the edifying camp were those who criticized the logical empiri-
cist approach, including Wittgenstein, Dewey, Austin, and Rorty himself.
As far as I can tell, the systematizers are not about to relinquish their
calling. Thus, a sympathetic reviewer of Rorty's book remarks:
If "epistemology" means the search for such foundation, then its end is in
sight, an end foreseen by Dewey. But if "epistemology" denotes an attempt to
understand the possibility and nature of the various kinds of knowledge and styles
of reasoning, then Plato, Locke, and Dewey are part of a persistent tradition that
has to do with one of the essential characteristics of our civilization. (Hacking 1980,
p. 586)
And Hilary Putnam, a contemporary of Rorty, offers a harsher indictment.

Putnam is particularly disturbed by Rorty' s assertion, from the perspective
of cultural relativism, that epistemology is just a particular hang-up of the
Western intellectual tradition:
Cultural relativism is a denial of the possibility of thinlcing ... [and a] sugges-

tion ... that philosophy ... is a silly enterprise. But the questions are deep and
it is the easy answers that are silly. Even seeing that relativism is inconsistent is,
if the knowledge is taken seriously, seeing something important about a deep
question. Philosophers are beginning to talk about the great issues again, and to feel
that something can be said about them, even if there are no grand or ultimate
solutions. There is an excitement in the air. And if I react to Professor Rorty's book
with a certain sharpness, it is because one more "deflationary" book, one more
book telling us that the deep issues aren't deep and the whole enterprise was a
mistake, is just what we don 'f need right now. (Putnam 1981, p. 236)
77
I I I THE CoGNITIVE SciENCES : HISTORICAL PERSPECTIVE
Putnam is adverting to the belief (or faith) among several philosophers

that epistemology is not dead. In fact, in light of cognitive-scientific events
occurring at mid-century, new life was breathed into epistemology.
Thanks especially to the invention and dissemination of the computer, but
also to techniques and findings in a host of allied fields, it is now possible
to revisit some classical issues of philosophy and to conceptualize mental
processes in new and fruitful ways.
Fresh Approaches to Epistemology
Functionalism
One major contributor to the discussion is Hilary Putnam himself. A
mathematically trained philosopher, Putnam has had a long-standing in-
terest in the nature of computers and their implications for thinking. As
he himself recalls, the invention of computing machines was an important
event in the philosophy of mind because it led to the idea of functional
organization (1973, p. 299). This functionalist idea challenges the assertion
that thinking and other "intelligent functions" need to be carried out by
means of the same specified machinery in order to reflect the same kinds
of process. To be sure, before the advent of computing machines, it might
have been tenable to hold that thinking can only occur in human beings,
or in entities with the kind of brain structure that we have. But computers
demonstrated that many of the processes we would once have termed
"thinking" can indeed be realized by mechanisms that are constituted of
entirely different components-transistors or vacuum tubes-rather than
nerves, blood, and tissues.
If there was to be an identity, it obviously could not reside in the
hardware but, as Putnam pointed out, might well occur in the software:
that is, both human beings and machines-and any other form of intelli-
gent life, from anteaters to Antipodeans-could be capable of realizing the
same kinds of program. Thus, the equation occurs at a much higher level
of abstraction-a level that focuses on the goals of a cognitive activity, the
means of processing at one's disposal, the steps that would have to be
taken, the evaluation of the steps, and kindred features.
Indeed, functionalism purported to address one of the most classic
philosophical dilemmas-the mind-body problem. Thought can indeed
occur in a physical apparatus and can be correlated with certain behavior
and yet not have to be identified with the precise class of activities that
78
happen to be produced (Matthews 1982). It is perfectly legitimate to talk
of mental events, to posit one mental event as causing another, and to do
so without taking a position on whether only brain events have the proper-
ties to define mental states. Of course, in human beings, mental events are
identical to physiological ones, but there is no need to reduce psychological
explanations to the neuro-physiological level. Rather one may counte-
nance a level of explanation that connects psychology to neurology, a
second level that connects it to social factors, and even a third free-
standing level, which is representational.
Intentional Systems
Building on Putnam's pioneering attempt to tackle the issues of
knowledge in an idiom that makes sense in the contemporary cognitive
milieu, certain philosophers in more recent times have put forth their own
analytic schemes. Daniel Dennett begins by distinguishing his own theo-
ries from earlier efforts (Dennett 1978). He does not wish to embrace a kind
of physicalism in which, for every type of mental event, there is a particu-
lar type of physical event in the brain. Nor does he want to adopt a
so-called Turing-machine functionalism, which would imply that all in-
dividuals have the same program. He thus searches for a level of explana-
tion where it is possible to talk about what two individuals or two entities
have in common psychologically without both being the realization of
some single Turing machine.
Enter the intentional system. This concept is designed to play a role
in the legitimation of mental descriptions parallel to the role played by the
abstract notion of a Turing machine in setting down the rules for interpret-
ing an artifact like a computer. Intentionality is seen as the mark of the
mental, and Dennett claims that every mental phenomenon can be de-
scribed in terms of intentional systems.
As Dennett phrases it, one may consider both human beings and a
computer and its programs as agents whose acts one is trying to explain.
Though one can talk about the computer in terms of its physical design
or its actual physical states, it often makes more sense to treat the machine
as if it were significantly like an intelligent human being. In other words,
one attributes rationality and purpose to an intentional system. And so, if,
for example, you are playing a game of chess with a computer, you think
of the program as having at its disposal certain goals, procedures, strate-
gies; and you try to outwit the program just as you would try to outwit
that more conventional kind of intentional system-another person.
The notion of an intentional system, Dennett argues, is relatively
uncluttered and unmetaphysical. It does not touch the issue of the compo-
79
sition, quality, morality, or divinity of any of its instantiations. Intentional

systems serve as a bridge from the common-sense world of persons and
actions to the non-intentional domain of the standard physical sciences.
Artificial intelligence is a handmaiden in this enterprise, because this disci-
pline builds a system that is in fact knowledgeable, and thus provides one
answer to the question of how knowledge is possible.
Intentional systems emerge at a certain level of complexity. One can
break down an intentional system into subsystems, each one of which is
itself viewable as an intentional system, and then break it down in terms
of even finer systems. These subsystems can be thought of as little homun-
culi which communicate with one another. Ultimately, however, one
wants to reach a level where these various homunculi are "discharged,"
where there is no need to adopt the intentionalist stance. As Dennett
puts it:
If one can get a team or committee of relatively ignorant, narrow-minded, blind

homunculi to produce the intelligent behavior of the whole, this is progress .
. . . Eventually this ... lands you with homunculi so stupid (all they have to do
is remember whether to say yes or no when asked) that they can be, as one says,
"replaced by a machine." One discharges fancy homunculi from one's scheme by
organizing armies of such idiots to do the work. (Dennett 1978, pp. 123-24)
Which touches upon the hedge in Dennett's formulation. While he

finds it useful to involve "intentional system talk" for the machines of
artificial intelligence, he asserts, at the same time, that this is just a manner
of speaking:
[T]he definition of intentional systems I have given does not say that inten-
tional systems really have beliefs and desires but that one can explain and predict
their behavior by ascribing beliefs and desires to them, and whether one calls what
one ascribes to the computer beliefs or belief-analogues or information complexes
or intentional whatnots makes no difference to the nature of the calculation one
makes on the basis of the ascriptions. (1978, p. 7)
But if this is only a manner of talking, what has been gained by this
apparently daring move? It is likely that Dennett, a student of Ryle' s, would
really rather dispense with this mentalistic line altogether and fall back on a
more trustworthy behaviorist mode of discussion. Still, Dennett concedes
that such discussion really is no longer adequate for scientific purposes. As
he comments in a critical piece on the behaviorist B. F. Skinner, "There is no
reason why intentional terms can't be used provisionally in the effort to
map out the function of the behavioral control system of men and animals
just as long as a way is found to' cash them out' by designing a mechanism to
function as specified" (Dennett 1978, p. 62).
80
In developing their notions, Putnam and Dennett have exemplified
the contribution of philosophy to cognitive science. Responding to the
various issues raised by the advent of the computer, while drawing on
long-standing discussion about the mind's relation to the body and an
agent's sense of purpose, these authorities have helped to clarify issues
raised in the contemporary science of mind. Still, I think it would be fair
to state that both of these philosophers have entered the terrain of cogni-
tive science with considerable caution. And, as we shall see in later chap-
ters, more vexing issues have been raised, by philosophical critics of artifi-
cial intelligence, such as John Searle and Hubert Dreyfus. For these
reasons, it is salutary to consider the work of Jerry A. Fodor, a full-scale
cognitivist-one philosopher who seemingly has no reservations whatso-
ever about the common fate of philosophers and empirical scientists inter-
ested in issues of mind.
The Complete Cognitivist: Jerry Fodor
To read Fodor (1975, 198lb) is to feel that the tables of epistemology have
been suddenly and radically turned: the "bad guys" have been propelled
to the position of heroes and some widely heralded common heroes have
been demoted. Following his one-time mentor and present colleague Noam
Chomsky, Fodor finds much to admire in the analysis put forth by Des-
cartes over three hundred years ago. For Fodor, the Cartesian tradition has
the merit that it recognized the existence of mental states and freely per-
mitted mental events to have causal power. Furthermore, it countenanced
the positing of innate ideas-informational content, mechanisms, or prin-
ciples with which the individual is born and which allow the individual
to make sense of experience. As part of his own version of epistemology,
Fodor is vociferously critical of the empiricist tradition. He feels that three
hundred years of empiricist efforts, from Hume to the logical positivists,
have resulted in failure. On Fodor's reading, both empiricists and rational-
ists had accepted that some concepts must be innate, but he sides with the
rationalists in their belief that the human being is born with the knowledge
of many concepts, which, at most, must simply be triggered by the environ-
ment. Fodor also excoriates the Rylean behaviorist position: he, in fact,
wants to claim that the kinds of explanations Ryle rejected are just those
that are needed. In Fodor's view, it is quite a different matter to explain
the circumstances under which an utterance is made from the reasons it
is made; and, in defiance of Ryle, it is part of Fodor's project to invade the
81
I I I THE CoGNITIVE SciENCES : HisTORICAL PERSPECTIVE
mental space: to figure out just why and how we make the statements we
make, given the mental equipment with which we are endowed.
But Fodor's rejection of much past philosophical work should by no
means be interpreted as an uncritical return to the tenets of Cartesianism.
First of all, he stresses that mentalism is not equivalent to dualism. One
can believe in the existence of mental states and in their causal efficacy
without believing that there are two substances-mind and matter-which
must somehow interact with one another. In this case, he is putting forth
a materialistic variant of mind, which nonetheless allows the existence of
mental causes and the interaction of mental states with one another.
Indeed, Fodor accepts much of the functionalist perspective as first
introduced by his teacher, Hilary Putnam. On this view, the psychological
constitution of a system depends not on its hardware (or its physical
realization) but on its software: thus, Martians can have pains, and com-
puters can have beliefs. Fodor goes on to embrace the general information-
processing approach of the cognitive sciences: it is in the manipulation of
symbols, or mental representations, that cognitive activities are realized
and, in fact, constituted. Consistent with many contemporary philoso-
phers of mind, and in direct opposition to those earlier philosophers who
saw mind as a mirror of nature, Fodor rejects resemblance as a property
of mental representation: the symbols of the mind are abstract entities
which need bear no configurational relationship to the entities they denote.
Finally, Fodor's functionalism in no sense reflects a reductionist ap-
proach. Intelligence and mental states can be realized in many systems and
entities, and there is no priority given to explanation in terms of the
biochemical or the neurological. Indeed, Fodor voices his strong reserva-
tions that the "natural kinds" of the nervous system will map in any
interesting way onto the "natural kinds" of psychological or mentalistic
explanations. If anything, the tie between mind and computer is likely to
be more intimate than the tie between mind and brain. Throughout, there
is a strong "intentionalist flavor" to the discussion: Fodor injects an idiom
of beliefs, desires, goals, and the like entering into the very center of
discussions about various kinds of cognitive systems.
Until this point it may seem that, while Fodor features a different cast
of heroes and villains, his approach is consistent with those of other
cognitively oriented epistemologists, such as Putnam and Dennett. And,
indeed, viewed from afar, and compared with skeptics like Rorty or Ryle,
these philosophers can all be seen as enthusiasts of the computational
metaphor in the world of cognitive processes. But Fodor has gone well
beyond his contemporaries in his willingness to think about what mental
representation might be like. Here his Cartesian allegiance becomes most
patent.
82
In a word, Fodor believes that there must be a language of thought.
If cognitive systems involve representations, if cognitive operations in-
volve the manipulation of symbol-like representations, then these repre-
sentations must exist somewhere and be manipulated in some way. Ac-
cordingly, Fodor feels that a commitment to attributing a representational
system to organisms must entail a characterization of this mentalistic sys-
tem. He asserts, "I am proposing ... to resurrect the traditional notion that
there is a 'language of thought' and that characterizing it is in good part
what a theory of the mind needs to do" (1975, p. 33). Fodor feels that
acknowledgment of some kind of medium, or language, within which
thinking takes place is an implicit part of nearly every contemporary
cognitive theory. It has become his burden to try to spell out what that
language might be.
And so to Fodor's long and highly stimulating essay The Language of
Thought (1975), seen by some commentators as a "great divide" in twen-
tieth-century philosophy (Piattelli-Palmarini 1983). In this work, Fodor
claims that the language of thought must be an extremely rich vehicle if
it is to carry out the many cognitive processes-perception, reasoning,
language-learning, and the like-of which human beings are capable. In-
deed, if mental processes are computational, there must be representations
upon which computations can be performed. Moreover, he puts forth his
radical view that this language of thought is innate: that people are born
with a full set of representations, onto which they can then map any new
forms of information that happen to emerge from their experiences in the
world. Furthermore, "the language of thought may be very [much] like a
natural language. It may be that the resources of the inner code are rather
directly represented in the resources of the codes we use for communica-
tion .... [This is] why natural languages are so easy to learn" (1975, p. 156).
Fodor's claim that individuals are born with knowledge of a language
-an innate language similar to a natural language-is astonishing, and it
is not certain just how seriously he means it to be taken. But his challenges
to competing accounts are deadly serious. Take, for example, his critique
of Jean Piaget's theory of concept acquisition, according to which the child
comes to possess new and more powerful concepts at each subsequent
stage of development. Fodor parades the difficulty he has in understanding
how one can ever learn a new concept unless one already has the ability
to hypothesize that concept-in which case, one already possesses it.
Shades of the Meno! Similarly, in his critique of the claim that complex
concepts are built out of simple ones, Fodor argues strenuously that all
attempts a!a Locke to identify the building blocks of larger concepts have
failed, and that this failure is putative evidence that this particular "em-
piricist" maneuver was ill conceived from the start. The fact that the
83
operations that individuals can carry out, even early in life, are highly
abstract, gives further weight to Fodor's (and Chomsky's) claims that the
initial intellectual apparatus with which individuals are equipped must be
sharply specified, constructed so as to expect particular experiences and
information. Even though the exact claims put forth by Fodor have failed
to persuade most of his colleagues, the issues he raises about the need for
some kind of "mentalese," and the kinds of constraints this mentalese may
have to exhibit, have proved difficult to undermine. Thus, his position
passes one of the most critical tests for any philosophical claim.
From Fodor's point of view, it is inadequate to conceive of the lan-
guage of thought as simply a formal medium of symbol manipulation (see
Stich 1983). There must be some way in which the contents of the world
are represented mentally, for we do not just think, we think about certain
things, and those certain things surely exist in the world. Yet at the same
time, Fodor expresses pessimism about the likelihood that we will ever be
able to understand how content is dealt with by our computational systems.
We may be restricted, as investigators, simply to describing the kinds of
"syntactic" operation that are carried out, while remaining ignorant of the
ways in which these operations refer to the external world so as to lead
to different mental concepts. As Fodor puts it, it may be impossible for the
cognitivist to do semantics, even though "to deny that mental operations
have access to semantic properties of mental representations is not to deny
that mental representations have semantic properties" (198lb, p. 244).
Fodor does fear that these properties may be inaccessible to scientific
investigation. The machine of our minds does not know what it is talking
about, and does not care about a semantic relation. While so-called natu-
ralistic psychology wants to be able to explain how we learn to know about
the particular things of the world and what they are like, computational
psychology-a psychology of formal mental operations-may be the only
kind of psychology we can ever hope to get. On such "content-blind"
sentiments the historically oriented student of philosophy can discern
links between Fodor's enterprise and that of Carnap and the logical empiri-
cists: indeed, Fodor even adopts Carnap's term methodological solipsism to
characterize a syntactic approach to cognition.
In sum, then, Fodor believes that any attempt to understand cognition
must involve a full-fledged embracing of a mentalistic point of view. He
believes that mental states really exist, that they can interact with one
another, and that it is possible to study them. Methods of study involve
the empirical methods of psychology, linguistics, and other cognitive
sciences; and chances of making advances on these issues are tied closely
to an informed collaboration among experts in these different areas.
Fodor's own guess is that those mental operations that occur in a relatively
84
rapid and automatic fashion-like the syntactic parsing of a sentence or the
detection of forms in the visual world-are most likely to be elucidated by
current cognitive scientific methods. He is correspondingly pessimistic that
those capacities involving sustained judgment and reasoning-such as the
development of scientific theories or the making of decisions about ordi-
nary life-will lend themselves to the kind of syntactic (or formal) analysis
for which cognitive science has proved suited.
It might be thought that such a reversion to Cartesian notions, such
bold talk of a language of thought, would prove anathema to many of the
authorities I have discussed. In particular, it might be thought that Richard
Rorty would see Fodor's move as exactly the kind of tack a sophisticated
cognitive researcher ought to avoid. It is surprising, therefore, but also
illuminating, that while Fodor's work is reviewed critically in Rorty's
book, it is not by any means dismissed.
Rorty, it will be recalled, saves his harshest criticism for those who
believe in the Mirror of Nature-in a mental apparatus that in some way
can tell about what the world is like and help us judge it in the most
accurate possible way-to give us "right opinions," as he might put it.
Rorty notes that Fodor is interested in constructing a mechanistic model
of mental processes and is willing to posit all kinds of internal states for
doing so, but that his ultimate criteria call only for the kinds of description
that may yield a fruitful theory of cognitive processing-an avowedly
psychological non-epistemological program like Dodwell's. Rorty stresses
that the question, "How do we recognize bottles?," is entirely different
from the question, "What is indubitably given to the mind, such as to serve
as an infallible touchstone for inference?" (1979, p. 245). And so Rorty and
Fodor concur on the importance of separating out two questions: how the
organism interacts with the world-a legitimate psychological question;
and whether the organism's views about the world are in fact true-the
traditional (and, to Rorty's mind, the untenable) question. We now see,
Rorty points out, "that Fodor's picture of the mind as a system of inner
representations has nothing to do with the image of the Mirror of Nature
I have been criticizing. The crucial point is that there is no way to raise the
skeptical question 'How well do the subject's internal representations rep-
resent reality?' about Fodor's 'language of thought'" (p. 246).
Whatever Fodor's and Rorty's particular points of agreement may be,
Fodor, Chomsky, and others in the cognitive camp clearly wear the mantle
of Rorty's systematizers: they believe that the traditional questions are
important, and they want to try to solve them as best they can, though
they may retain rather more skepticism about what is possible than did the
great systematizers of a few centuries past. Rorty, on the other hand, as
a self-styled edifier believes that philosophy has, at most, puzzles, and that
85
I I I THE CoGNITIVE SciENCEs : HISTORICAL PERSPECTIVE
the whole agenda of epistemological questions had better be scuttled. And

however little he criticizes the kind of psychology in which Dodwell or
Fodor are engaged, he apparently thinks that psychology has thus far not
accomplished much, and believes that eventually only neurological and
humanistic approaches to mental phenomena will be left (Rorty 1982a).
The very fact that Fodor and Chomsky revert to the Cartesian tradition
and see themselves as providing answers relevant to the rationalist and
empirical disputes exemplifies for me the deep gap between these two
strands of current philosophical thinking.
Conclusion: The Dialectic Role of Philosophy
When it comes to our major issue of philosophical content-the question

whether the empiricist or the rationalist account has carried the day-the
pendulum has swung back and forth. For a brief period around 1800, it
appeared that Kant's synthesis had solved the problems once and for all;
but 130 years later, in the height of the behaviorist era, the empiricists were
in the ascendancy. Now, thanks to the cognitive revolution, and in the
wake of the influential writings of theorists like Fodor, rationalism is being
taken more seriously than it has been for decades. It is probably fair to say
that most cognitive scientists no longer care about which perspective
"wins" in this debate, and that the purpose of the Fodor critique is less to
resurrect Cartesian rationalism than to demonstrate the bankruptcy of the
empiricist position. Still, to the extent that cognitive science continues to
gain adherents, the rationalist position will emerge as essentially more
credible. The primacy of the knowledgeable subject-one who acquires
knowledge only by virtue of prior cognitive structuring (if not innate
ideas!)-is now accepted widely. In that sense at least, the trend of philoso-
phy is toward a rationalistic stance, one supported by empirical work in
several disciplines.
The role of philosophy as a discipline in its own right-and as a
continuing valid contributor to the cognitive sciences-is vexed. Several
reservations have been raised about philosophy from both within and
without the discipline. As we have seen, critics of logical empiricism-
most especially Richard Rorty-question the need for philosophy, at least
in its epistemological guise. From their perspective, the questions that once
motivated philosophers-from Plato through Kant-were actually mis-
conceived and are not susceptible to the kind of systematic answers their
predecessors had sought. It would be preferable for philosophers to aban-
86
don these issues-leaving them to empirical scientists to work out if they
so chopse-and instead become informed commentators about culture-
which in the case of our culture would include commenting on the great
philosophers of the past. A different argument is put forth by researchers
in the mainstream of cognitive science, for example, in artificial intelli-
gence. From their perspective, once one has provided computational ac-
counts of knowledge, understanding, representation, and the like, the need
for philosophical analyses will evaporate. After all, philosophy had once
helped set the agenda for physics; but now that physics has made such
tremendous strides, few physicists any longer care about the musings of
philosophers (Holton 1984).
I am not convinced by either of these critiques. On my analysis,
philosophy has been crucial from the first in formulating a set of questions
worthy of study and in monitoring the course of these questions over the
millennia. There is progress in philosophy-not perhaps as linear in the
physical sciences-but clear progress nonetheless. This progress comes
about because of the debates that take place among philosophers-for
example, when Locke criticized certain concepts of Descartes, arid Berkeley
and Hume in tum criticized Locke; and it occurs with equal impact through
interaction between philosophers and empirical scientists, such as those
discussions in which physicists and logical empiricists were involved in the
early part of this century.
I suggest, then, that philosophy participates in the disciplinary matrix
by virtue of its dialectical role: a dialectic within the discipline and a
dialectic between the analysis put forth by philosophers, on the one hand,
and the empirical findings and theories put forth by scientists, on the other.
This has happened dramatically in recent years. Just at the time when
philosophy seemed at low ebb, when the program of logical empiricism has
been thoroughly discredited, the invention of the computer and the begin-
ning of cognitive science suddenly underscored the need for sophisticated
analysis. It was thinkers acquainted with the long-standing philosophical
tradition-with Kant's notions about representations, Descartes's claims
about the mind-body problem, Locke's skepticism about innate ideas-
who could bring to bear the appropriate conceptual framework and then
revise it in the light of new scientific discoveries.
All this is not to claim that the traditional questions of philosophy
have inherent superiority. While Rorty's critique of traditional epistemol-
ogy seems like "overkill" to me, the kinds of questions pondered by
philosophers have changed over the millennia: some issues, such as the
nature of visual perception, which exercised the Greeks, become the exclu-
sive province of empirical science; some issues, such as the nature of "raw"
sensory experience, come to be seen as ill conceived; and some issues, like
87
the nature of intention and purpose, gain new urgency because of inven-
tions like the computer.
I see the invention of cognitive science as a wonderful stimulus for
philosophy, on the one hand, and philosophy as an indispensable hand-
maiden for the empirical scientists, on the other. Philosophy enables us to
define fundamental cognitive scientific questions in a coherent way, and
assures the proper integration of work in disparate fields. But, by the same
token, philosophy must attend assiduously to empirical findings in order
to avoid becoming a barren discipline or one irrelevant to scientific work.
It is thus fitting that the field of philosophy, whose initial agenda helped
to stimulate the rise of cognitive science, has been fueled by that new
discipline, even as philosophy can, in tum, help to inform and interpret
work spawned by its recent intellectual offspring.
Hilary Putnam, a veteran of many of these discussions, has reflected
on the role of philosophy in the contemporary scientific scene. His prudent
comments take seriously the various critiques of "grand" philosophy,
while recognizing the important role philosophy should continue to play
in discussions of new scientific endeavors:
I have not attempted ... to put forward any grand view of the nature of
philosophy; nor do I have any such grand view to put forth if I would. It will be
obvious that I do not agree with those who see philosophy as the history of
"howlers" and progress in philosophy as the debunking of howlers. It will also be
obvious that I do not agree with those who see philosophy as the enterprise of
putting forward R priori truths about the real world .... I see philosophy as a field
which has certain central questions, for example, the relation between thought and
reality .... It seems obvious that in dealing with these questions philosophers have
formulated rival research programs, that they have put forward general hypothe-
ses, and that philosophers within each major research program have modified their
hypotheses by trial and error, even if they sometimes refuse to admit that that is
what they are doing. To that extent philosophy is a "science." To argue about
whether philosophy is a science in any more serious sense seems to me to be hardly
a useful occupation .... It does not seem to me important to decide whether science
is philosophy or philosophy is science as long as one has a conception of both that
makes both essential to a responsible view of the real world and of man's place in
it. (Putnam 1975R, p. xvii)
88
5
Psychology:
The Wedding of
Methods to Substance
Three Pivotal Lines of Research from the 1950s
George Miller's Magic Number 7

In 1956, George Miller published, in the Psychological Review, an artfully
presented essay-"The Magical Number Seven, Plus or Minus Two: Some
Limits on Our Capacity for Processing Information." More of a synthesis
than a report of a crucial experiment or a presentation of a formal theory,
Miller's opening statement was backed by considerable empirical evidence:
"My problem is that I have been persecuted by an integer. For seven years
this number has followed me around, has intruded in my most private
data, and has assaulted me from the pages of our most public journals"
(p. 81). Miller showed that the individual's ability to make absolute dis-
tinctions among stimuli, to distinguish phonemes from one another, to
estimate numbers accurately, and to remember a number of discrete items
all seemed to undergo a crucial change at about the level of seven items.
Below that number, individuals could readily handle such tasks: above it,
individuals were likely to fail. Nor did this discontinuity seem accidental.
In Miller's view:
89
There seems to be some limitation built into us either by learning or by the

design of our nervous systems, a limit that keeps our channel capacities in this
general range. On the basis of the present evidence it seems safe to say that we
possess a finite and rather small capacity for making such ... Uudgments] and that
this capacity does not vary a great deal from one simple sensory attribute to
another. (P. 86)
Humans, however-as Miller reassured his audience-have ways of

getting around these constraints. Processing or coding entities in terms of
their various dimensions can enlarge the number of elements that can be
distinguished from one another. One can chunk, or group together, a
number of elements (for example, a set of numbers or letters) and then
treat the assemblage as a unit. One can make relative rather than absolute
judgments. The capacity to recode information into language and to re-
member this more abstract symbolism is especially important. As Miller
notes, "This kind of linguistic recoding that people do, seems to me to be
the very lifeblood of the thought processes" (1956, p. 95). Indeed, he once
described the potential efficiency of this recoding process in this manner,
"To use a rather farfetched analogy, it is as if we had to carry all our money
in a purse that could contain only seven coins. It doesn't matter to the
purse, however, whether these coins are pennies or silver dollars" (Miller,
Galanter, and Pribram 1960, p. 132).
Why did this apparently simple point have a decidedly major impact
within cognitively oriented communities? First Miller's essay brought to-
gether a large amount of hitherto dispersed data and suggested that they
pointed to a common conclusion. A valuable synthesis, to begin with.
Second, it suggested that the number 7 was no mere accident: it designated
genuine limitations in human information-processing capacities. While
such "built-in" limitations might be anathema to radical empiricists, they
helped signal a shift toward exploring the nature and structure of a central
cognitive processing mechanism. The point about strict processing limits,
not coincidentally, was made in terms of the theory of information, which
Miller explained early in the paper: thus, he introduced a method whereby
researchers could examine other sensory modalities or tasks and ascertain
whether this apparent limitation in fact obtains. (Much ensuing contro-
versy centered on how to make such translations and whether, when made,
they did in fact yield the magic number 7.) Third, as indicated, the message
in the paper was not without hope, for Miller indicated ways by which
humans ingeniously transcend this limitation.
There may have been another reason for the impact of this essay.
Psychologists had been trying for approximately a century to discover the
basic laws of the human mental system. Many promising avenues had
been launched, but most of them-including, most recently, the behavior-
90
Psychology: The Wedding of Methods to Substance
ist approach-had eventually foundered. In recent years, the most exciting
new work in the human sciences had come from two connected areas:
information theory, which posited principles of transmission applicable to
any kind of channel; and computer science, which now featured machines
engaged in symbol manipulation. Miller was holding out hope of marriage
between the quantities of data collected by psychologists over the years
and the rigorous new approaches of the engineering-oriented scientists.
The result might be a genuine science of psychology with its own set of
immutable laws. No one thought to question whether all contents, or bits,
can in fact be treated (and then counted) as equivalent.
The British Approach to the Processing of Information

Just as Miller and his colleagues were applying concepts from commu-
nication science to psychology, a parallel movement was getting under way
in Great Britain. This movement grew directly out of the applied psycho-
logical work that had been carried out in Britain during the Second World
War when psychologists joined other scientists in efforts to crack enemy
codes, understand night vision, plan air-raid alerts, assist in spotting of
enemy aircraft, and meet sundry other needs of war. Two men who had
been involved in this applied effort were Colin Cherry and Donald Broad-
bent; their studies in the 1950s inspired the British approach to informa-
tion-processing psychology.
A follower of information theory, Cherry (1953) focused on the capac-
ities of individuals to attend to and obtain information from noisy chan-
nels. He instructed subjects to follow a message, delivered to one ear, by
the method of shadowing: that is, by repeating each word as soon after its
initial presentation as possible. Cherry found subjects unable to report
much of what had come into the opposite (unattended) ear. More pre-
cisely, they could report gross characteristics of the signal, such as whether
it was music or speech, but not shifts of content or tongue. Broadbent
(1954) refined this procedure by presenting sets of digits simultaneously,
in strings of three, to both ears. He found that subjects had the easiest time
and achieved the highest score when they reported all the digits presented
to one ear first, and then all the digits presented (at the same time) to the
other ear.
For our purposes, the important part of the Broadbent-Cherry work
is the model of human thought processes to which it gave rise. The model
conformed to the tradition of British empiricism. It began with information
taken in from the senses, but focused on a new and important feature: that
the individual has a limited capacity for the intake and storage of informa-
tion. (In this stress on limits of information processing, the model was
91
intimately linked to George Miller's studies of the magic number 7.) There
was an important added twist: rather than simply speaking of structural
limits in a static way, the British researchers sought to determine precisely
what happens to this information from the moment one first apprehends it.
Given this "engineering" approach, it became a natural step to draw a flow
chart of what happens, as the perceptual system operates upon new infor-
mation. In fact, according to a recent textbook, Broadbent was the first
psychologist in modern times to describe cognitive functioning with a flow
chart (Lachman, Lachman, and Butterfield 1979, p. 188).
System for Varying

Output Until Some
Input Is Secured
Selective Limited Capacity

Filter Channel (p System)
Store of Conditional
Probabilities of
Past Events
One of the first information-processing diagrams.

SouRcE: From D. E. Broadbent, Perception and Communication (Elmsford, N.Y.: Pergamon
Press, 1958). Reproduced by permission.
What was this early flow chart like? It featured information coming
in through the senses, being placed in a short-term store, and then being
selectively filtered before entering into a limited-capacity perceptual sys-
tem. While a sense organ can take in a lot of information in parallel and
retain it momentarily, the job of the selective filter is to block unwanted
messages, thereby letting in only those that merit additional analysis. A
further property of the selective filter is that it can be so tuned that, at any
one time, it allows in only those messages that fulfill certain requirements.
The buffer can hold unanalyzed information briefly and thus allows one
to report the contents of the second ear, after having spewed out the three
digits apprehended by the first ear. According to Broadbent's early model,
only information that becomes conscious-that passes through the limit-
92
ed-capacity channel-can enter into long-term memory and thus become
part of active knowledge. Information present on an unattended channel,
or in an unattended signal, is assumed to decay in a few seconds and to
receive no processing beyond the initial "pre-attentive" analysis.
While few, if any, investigators believed that perception or thinking
takes place simultaneously or without a series of steps, the option of
tracing the stages of information processing had rarely been followed
before Broadbent's time. Again, this option became a probability when
communication engineering began to impinge on the perceptual and atten-
tional issues that had long interested psychologists. But the model of "flow
charting" put forth by Broadbent and his associates and their evidence
relevant to specific stages of information processing opened up many pro-
ductive possibilities. One could now examine the temporal dimensions of
diverse psychological processes, and avid experimenters lost little time in
pursuing just that course. The lack of attention to the particular content
being processed, or to the lcinds of transformation imposed, did not trouble
those excited by the Broadbent-Cherry demonstrations.
Jerome Bruner s Strategic Approach

With the collaboration of Jacqueline Goodnow and George Austin,
Jerome Bruner published in 1956 a book called A Study of Thinlcing. This
volume grew out of the Cognition Project, which Bruner had been direct-
ing for several years at Harvard. The subject, well known to psychologists,
was classification, categorization, or (as it was commonly called in the
trade) concept formation or concept acquisition. And the problem was a
classic one: How does a person, confronted with a set of elements, come
to group them together reliably into categories-be they all chairs, all
atoms, or all large blue triangles?
As psychologists, Bruner and his colleagues followed the tradition of
examining abstract forms of categorization, such as those involved in figur-
ing out which of a set of cards, each featuring a different geometric form,
belongs to a particular category. The experimenter would target a concept
-say, the class of all cards with one red figure, the class of all cards
featuring red squares, or, when being particularly diabolical, the class of
all cards containing two figures and/ or circles. The subject was exposed to
one card at a time, asked in each case whether that card belonged to the
preordained concept, and then told whether his response was correct. The
subject's task, of course, was to figure out the properties of the targeted
concept, so as to be able to select all, and only, those cards that exhibited
its defining features.
While superficially similar to work carried out in years past, Bruner's
93
approach actually diverged from that undertaken by earlier students of

categorization. First of all, rather than treating subjects as deaf and mute
animals, Bruner and his colleagues simply told the subjects what to do and
relied heavily on their comments as an aid in analyzing the results. Flying
in the face of established behaviorist methodology, the subjects were
treated as active, constructive problem solvers, rather than as simple
reactors to the stimuli presented to them. Their introspections actually
mattered.
Reflecting the information-theoretical winds of the day, Bruner and
his associates had begun by looking at the bits of information assimilated
in subjects' encounters with such simple stimuli. But in another departure
from standard operating procedures, they ended up analyzing the informa-
tional properties of long sequences of acts called "strategies." It had turned
out that the best way to account for individual performances was in terms
of these overall patterns of responding over many trials rather than of
particular responses to a particular stimulus configuration. This was the
most iconoclastic-and the most influential-aspect of the work.
The researchers went on to consider what each strategy accomplished
in light of the goals of the organism, such as minimizing risk or conserving
storage capacity. Singled out were the strategy of successive scanning, where
the subject has a single hypothesis (like all red objects) and limits his
choices to those instances that directly test this hypothesis; the approach
of conservative focusing, where one finds a positive instance and then makes
a sequence of choices, each of which alters but a single attribute value of
the first "focus" card and tests whether the change yields a positive or a
negative instance; and the tack of focus gambling, where the subject uses a
positive instance as a focus but then takes the calculated risk of changing
more than one attribute at a time. Focus gambling offers the possibility of
attaining a concept far more rapidly than does conservative focusing, but
also may necessitate extra trials if one's choices happen to be unrevealing.
Conservative focusing is the most foolproof method since it limits the
burden on memory and allows steady progress toward a solution. How-
ever, in case of a time limit or some other pressure, the respondent may
adopt the riskier course of changing several attributes simultaneously.
Perhaps not surprisingly for a work published in the pivotal year of
1956, the Bruner book presented itself as an innovation in psychology. The
authors comment:
the past few years have witnessed a notable increase in interest in and investi-
gation of the cognitive processes.... It has resulted from a recognition of the
complex processes that mediate between the classical "stimuli" and "responses"
out of which stimulus-response learning theories hoped to fashion a psychology
94
that would by-pass anything smacking of the "mental." The impeccable peripher-
alism of such theories could not last.... One might do well to have a closer look
at these intervening "cognitive maps." (P. vii)
The importance of Bruner's book was signaled by praise from the distin-
guished physicist J. Robert Oppenheimer: 'A Study of Thinking has in many
ways the .flavor of the opening of a new science.... The book has a unity
of view and a fervor of conviction which makes it point to the future"
(quoted in Bruner 1983, p. 121). The possibility that the use of such
artificial concepts might invalidate the findings was far from anyone's
mind at that time.
The Program of Cognitive Psychology

The lines of research inaugurated by Miller, Broadbent and Cherry,
and Bruner, energized the psychology of the late 1950s and the 1960s. In
the face of the artificially tough strictures that had been imposed by
behaviorism on issues of cognition, these young psychologists were willing
to introduce notions that had long been ruled "out of court." Talk of
built-in limitations to the amount of information that could be taken in,
attempts to trace the steps involved in processing such information, and
positing of overall strategies employed to solve a problem-all of these
signaled a greater willingness to discuss issues of mind directly, without
attempting to explain them away in terms of long series of publicly verifia-
ble stimuli and responses.
Such a change cannot be attributed to a single factor, but it is clear
that the advent of the computer, as well as the information-theoretical
language by which it was commonly characterized, helped to legitimize
such approaches. No longer were psychologists restricted in their explana-
tory accounts to events that could either be imposed on a subject or
observed in one's behavior; psychologists were now willing to consider the
representation of information within the mind. To be sure, this willingness
to deal with mental representation took different forms in the writings of
different psychologists. Miller looked at the structural properties and limi-
tations built into the representational system; Broadbent and Cherry ex-
amined the transformations on the information as it came in from the
senses and was stored in memory; Bruner attributed to subjects a variety
of approaches or strategies which governed performance on a task. While
the issues being probed-memory for isolated units, processing of words
or tones, sorting of concepts-were scarcely new to psychology, the pros-
pect of applying concepts from information theory, of building on the
model of a computer, of countenancing various forms of mental represen-
95
II I THE CoGNITivE SciENCES : HISTORICAL PERSPECTivE
tation, and of allowing subjects to use their full reflective powers was
bracing and freeing.
Psychology is a discipline central to any study of cognition. Yet it is
also a difficult discipline to pursue and one where genuine progress has
been hard to achieve. Nearly every conceivable element is relevant to a
subject's performance, and few issues having to do with human nature and
behavior can be excluded from the laboratory a priori. Thus, choosing a
problem, and screening out all competing ones, becomes an especially
vexing task.
Psychology also poses special problems for the historian of cognitive
science-a problem in no way minimized when the historian is also a
psychologist. It is an enormous field-there are many more psychologists
than there are representatives of the other fields-and there are conse-
quently more programs of research to survey. While it is oversimplification
to organize any field around one or two themes, it is especially difficult to
select key issues in psychology. Should one, thus, pay attention to the
particular content of information (auditory or visual, musical or linguistic)
or instead treat all contents as if they were interchangeable? Does one
approach research in order to illuminate those processes that are true of all
individuals, or look instead at pertinent individual differences-child ver-
sus adult, male versus female, naive versus trained in experimental tasks?
Does one examine behavior in its natural context or try to strip away all
everyday accouterments and resort to artificial laboratory conditions?
Does one assume that the individual approaches tasks by building up
larger elements of meaning from small, isolated units? Or does it make
more sense to assume that one comes to tasks with general strategies or
scripts, which one simply imposes upon a task, irrespective of its particular
dimensions, details, and demands?
I have elected to organize this chapter in terms of a distinction that
touches upon some of the aforementioned ones but is perhaps better
phrased in somewhat different terms: that is, the distinction between
molecular or small-scale units of analysis and molar or large-scale units of
analysis. For reasons of scientific strategy or simple personal preference, it
seems possible to classify most psychological research programs along this
dimension. Some programs, such as those of traditional psychophysics and
contemporary information processing, show a penchant for small-scale
units (bits, individual percepts, single associations examined in brief peri-
ods of time) on the assumption that a thorough understanding of these
elementary units and processes is the surest path toward the ultimate
explanation of complex units and entities. A contrasting faith is found
among proponents of the molar approach-those who look at large-scale
problems tackled over a long period of time and invoke analytic concepts
96
like schemas, frames, or strategies. According to these researchers, these
large-scale properties are most salient in human cognition and thus serve
as a logical point of departure. Why gamble that an elementaristic ap-
proach will eventually yield larger units, when one has the option of
beginning instead with these larger units, which seem closer to the data and
the experiences of everyday life?
The contrast between molecular and molar approaches resembles, but
is by no means identical to, the distinction between the top-down and the
bottom-up approaches. In a top-down approach, which has rationalist over-
tones, the subject is assumed to bring to the task his own schemes, strate-
gies, or frames which strongly color his performance. In a bottom-up
approach, more allied to the empiricist camp, the actual details of a focal
task or situation are assumed to exert primary influence on a subject's
performance. In what follows, I shall often identify molar with "top-down"
and molecular with "bottom-up"-not because each is logically bound to
the other but because they often, and perhaps typically, occur together.
Like all dichotomies, this one is easily exaggerated, with subsequent
distortion of the field. Nearly all psychologists have some sympathy for
each tack, and many move from a molecular to a molar approach (and back
again). For example, George Miller favored a molecular hat when ponder-
ing the number 7 but readily shifted to a molar one when discussing plans
and goals in the 1960 volume. Indeed, when the computer is used as a
model, it is equally justifiable to focus on the most molecular level (indi-
vidual bits, symbols, "on-off" circuits) or the most high-level program-
ming concepts (goals, means, and routines). Also, one can embrace a mo-
lecular (or a molar) approach for different reasons: some psychologists
begin with a molecular approach in the hope of being able to adapt their
methods to molar entities; while others believe that ultimately all behavior
can be reduced to, and explained by, molecular entities. Thus, in embracing
this dichotomy, I seek to convey an ongoing tension or struggle for the soul
of psychology-not to label two bins into which one can readily and
reliably sort all experiments, concepts, and psychologists.
Two other trends must be mentioned as well in any thumbnail
sketch of psychology's first one hundred years. The first trend is the
increasing splintering of the field. The American Psychological Associa-
tion alone has over fifty thousand members (including several thousand
active researchers) who spread themselves over forty divisions and sev-
eral hundred special interest groups, many of which are completely igno-
rant of what is going on elsewhere in their association and their disci-
pline. In this climate, efforts to find unifying concepts are vital but by no
means easy to sustain.
The second is the trend toward methodological perfection. With the
97
I I I THE CoGNITIVE SciENCEs : HISTORICAL PERSPECTIVE
passage of time, the invention of new instruments, and the increased

sophistication of statistical techniques, the design of individual studies and
sets of studies has become increasingly elegant. No one would deplore this
situation, but the question must be raised whether (in comparison, say,
with molecular biology) this increasing methodological sophistication has
deepened our understanding of psychological phenomena. In other words,
have we obtained more penetrating knowledge about human cognition, or
are we simply using more convincing experimental demonstrations to
reaffirm knowledge that has long since been established?
In my view, its methodological sophistication is among psychology's
proudest achievements, but it is an attainment that has not yet been fully
integrated with the substance of the field. Many of psychology's most
important issues need to be approached from a molar perspective and
entail a top-down perspective. And yet the most rigorous psychological
methods often are not appropriate for these large-scale issues. As I see it,
the challenge currently faced by psychology involves a wedding of its
sophisticated methodological armamentarium to issues and questions of
unmistakable consequence. Since many of these questions arise from the
philosophical tradition out of which psychology directly grew, it is appro-
priate to begin this survey of the course of cognitive psychology with a
brief consideration of the Kantian legacy to nineteenth-century psycho-
logical studies.
Scientific Psychology in the Nineteenth Century
Coping with the Kantian Legacy

Immanuel Kant had severe doubts about the possibility of a science
of psychology-a skepticism that, as I noted in the last chapter, arose from
several causes. He believed that a science has to apply mathematical laws
to empirical data, and that such data have to be collected in real experi-
ments, but because psychology deals with elements that putatively have
no spatial dimensions-pure thoughts-such experimentation was not
possible. A second problem was that psychology would have to consider
the instrument of knowing-the self; but it is not possible for the self to
examine its own workings, let alone to do so in a disinterested way. There
was, in addition, the problem of the level of abstraction. To conduct
scientific research, one has to be able to strip away accidental factors so as
to focus on the variables crucial to a theory-a radical manipulation of the
98
subject matter difficult, if not impossible, to bring to bear on complex and
all-pervasive human interaction.
Kant's most severe objections derived from his overall conception of
the domain of knowledge. In his "foundational" view, it was the province
of philosophy to lay out the nature of thinking-to chart the relations
among the various sciences and to designate their foundations and limita-
tions; psychology was seen as a second-rate poacher upon such a program.
Psychology should be content to look at the social and historical contexts
in which thinking occurs but should not attempt to crack the nature of
thought itself.
Such was the authority of Kant-and the surface persuasiveness of his
arguments-that many scholars of his time shied away from the empirical
investigation of psychological issues. Fortunately, however, at least some
individuals were spurred by Kant's skepticism to seek a more positive role
for psychology. According to the historian of psychology, David Leary
(1978), a trio of German thinkers-Jakob Friedrich Fries, Johann Friedrich
Herbart, and Friedrich Eduard Beneke-directed their energies, in the early
part of the nineteenth century, to the conceptualization of a scientific
psychology. Each of these scholars believed that mental processes could be
measured by experiment and that studies could be carried out that actually
provided information on the operations of the mind. In particular, Herbart
claimed that ideas exhibit the variables of time, intensity, and quality, and
that one should be able to measure quantitatively each of these aspects of
ideation and even to write equations mapping their relations one onto
another. While the work of Her bart and his colleagues was basically re-
stricted to armchair speculation, these scholars kept alive the possibility of
a scientific psychology during an era where Kant's strictures remained a
formidable obstacle.
By the middle of the nineteenth century, scientists came to have fewer
reservations about empirical investigations bearing directly on psychologi-
cal issues. Less under the shadow of Kant and other philosophers, more
willing to carry out experiments and simply see what would turn up, these
scholars directly anticipated the founding of scientific psychology toward
the end of the century and set up ripples that can still be detected in the
laboratories of today. Moreover, in each case, they made specific contribu-
tions that remain pertinent to psychological discussions.
Laying the Groundwork: Helmholtz, Fechner, Donders, and Brenfano

Of the many empirically oriented researchers who displayed an inter-
est in psychological issues, probably the most outstanding was the German
physicist and physiologist Hermann von Helmholtz. Helmholtz wanted to
99
show that much of what Kant had speculated about could be subjected to
empirical study. Skeptical of various competing claims about the astound-
ing speed of thought, Helmholtz actually undertook to measure the length
of time it takes to transmit impulses along a nerve. Using an ingenious
instrument, adapted from his laboratory galvanometer, Helmholtz was
able to measure the time that it took for a nerve impulse to pass through
the severed leg of a frog. He then succeeded in adapting the method for
human subjects, who were asked to respond by pushing a button when-
ever a stimulus was applied to their legs. This allowed him to ascertain the
speed of the impulse in human sensory nerves: it turned out to be between
165 and 330 feet per second (Fancher 1979, pp. 100-101). Human behav-
ioral reactions could be measured, after all.
Having successfully challenged the belief that thought is essentially
instantaneous or inherently unmeasurable, Helmholtz went on to call into
question Kant's beliefs in innate ideas of space. According to Helmholtz's
rival position, individuals build up knowledge of space, just as they con-
struct the facts of their physical world. By having subjects wear prisms
which distorted their world, Helmholtz showed that individuals could
readily adjust or adapt to these distortions and were soon seeing the world
again in essentially undistorted fashion. In a complementary investigation,
he studied individuals who had once been blind but could now see, and
documented the considerable time that elapsed before they learned to
perceive the world of objects in the manner of the sighted.
From such demonstrations, Helmholtz developed the still-influential
idea of unconscious inference: rather than simply reading off percepts from the
world of external stimulation, we unconsciously draw on our past knowl-
edge in order to effect accurate interpretations of what we perceive. The
experience of one's past perception is unconsciously added to one's present
reaction to a stimulus-as happens when, defying the evidence of the
senses, one succumbs to an optical illusion. Helmholtz used the word
inference deliberately. He believed that the visual system was implicitly
reasoning about its experiences: for example, in order to figure out the
actual sizes of objects, the visual system had to make inferences based on
the images formed on an individual's retina. Unlike syllogistic inferences,
of course, these processes occurred without conscious awareness.
In forging ahead with his research, and in introducing certain concep-
tual distinctions, Helmholtz made three major contributions. First of all,
he indicated that Kant's philosophical dicta did not have absolute validity:
it was indeed possible to illuminate aspects of human mental functioning
in an empirical fashion. Second, Helmholtz cleared places for molecular
forms of analysis (the speed of an impulse traveling along a nerve fiber)
as well as molar investigations (the ways in which complex spatial arrays
100
are seen under both normal and distorted conditions). Finally, by stressing
the perceiving subject's contribution to perception, Helmholtz became an
early contributor to the ideology of cognitive science.
It now became possible for scientists of a less overpowering status to
contribute to the incipient science of psychology. The pioneering psycho-
physicist Gustav Fechner (1912) was able to show that, within limits, the
intensity of a perceived sensation varies as a logarithmic function of objec-
tive features of the stimulus. Thus it seemed that the most personal aspects
of a psychological experience-how loud a sound seems, how bright a light
appears, how sweet an object tastes-could bear a quantitative relation to
a measurable characteristic of an object in the world.
Building on Helmholtz's documentation of the speed of neural events,
F. C. Donders proposed in 1868 that one could also measure the time that
it takes for higher mental operations to be carried out. For instance, in order
to measure how long it takes to effect discrimination between two stimuli,
one would subtract the time that it takes to detect a single event in isola-
tion from the time it takes to respond to only a single event when two
events are presented. By subtracting the time needed to detect a single
event in isolation from the time needed to respond when given two stimuli,
one could infer the exact time required for the operaHon of discrimination. Like
Fechner, Donders worked with relatively molecular materials: but, in prin-
ciple, their methods could be applied to molar tasks.
While Helmholtz, Fechner, and Donders were all concerned to dem-
onstrate that psychological issues could yield quantitative results in the
experimental laboratory, a priest-philosopher named Franz Brentano was
approaching questions of psychology from a different standpoint. Bren-
tano's writing was directed against the molecular notion that one can break
psychology down into elements and examine the elements of experiences
or consciousness in isolation, as well as the notion that one can conceive
of thought processes or consciousness in a purely mechanistic way. For
Brentano, psychology starts with the mind-an active, creative entity
which has intentions, for it implies and demands an object. The true
subject matter of psychology is the mental act-such as judging, sensing,
imagining, or hearing, each of which reflects a sense of direction and
purpose. One cannot simply see; one must see something; and the act of
seeing something is psychological or mental.
Given this viewpoint, the task of empirical psychology is to study the
mind of the agent at work, dealing with objects, purposes, and goals.
Brentano stressed the phenomenological aspects of the psychological en-
deavor: one cannot conceive of thoughts and judgment, let alone study
them, except by taking into account one's inner phenomenal experience.
And this can be accessed not by prompted introspection-for one cannot
101
II I THE COGNITIVE SCIENCES : HISTORICAL PERSPECTIVE
observe at the same time that one experiences-but rather by simple phe-
nomenal experience of one's inner mental life (Brentano 1874).
The words of Brentano would be stilled temporarily during the first
years of the discipline of psychology, when an obsession with studying the
elements of sensations, and the basic constituent mental states, carne to the
fore. The laboratories of Leipzig had little place for such high-flown senti-
ments as act, purpose, intention, and phenomenal experience. Yet the
kinds of top-down or molar concern that Brentano had raised could not
forever be ignored. They were to re-emerge, in different forms, in the
rebellion in Wiirzburg, in the platform of the Gestalt psychologist, and,
once again, in the view of the computer as an agent with plans, intentions,
and goals. In not a few ways, Brentano has had profound influence on the
course of psychology and has affected many experimenters who have
never heard his name.
Wundt s Program
In a sense, psychology was well under way by the latter part of the
nineteenth century. With the pioneering work of Helmholtz, and the con-
siderable efforts by Donders, Fechner, and Brentano, who devoted them-
selves chiefly to psychological matters, there was already a minor industry
of psychological thinking and research. And yet, as I shall note repeatedly
in this history, the flowering of a discipline depends heavily upon the
founding of institutions and organizations. The efforts of the earlier nine-
teenth-century pioneers might never have coalesced and become cumula-
tive had it not been for Wilhelm Wundt, who took it upon himself in the
latter part of the century to establish psychology as a separate experimen-
tal discipline. More than any other individual-perhaps than any collec-
tion of individuals-Wundt is responsible for the emergence of psychol-
ogy as a separate scientific discipline with its own methods, programs, and
institutions. Psychology has grown from its original establishment in a
single university-Leipzig in 1879-to a field with representatives in virtu-
ally every institution of higher learning. Its spectacular success can be
attributed in no small measure to the labors-if not always to the ideas-
of Wundt (Boring 1950; Fancher 1979; Watson 1979).
According to Wundt, physics studies the objects of the external world:
while this investigation is necessarily mediated by experience, physics is
still not the study of experience itself. Psychology, in contrast, is the study
of conscious experience as experience. It must be approached through
internal observation, through introspection. While all individuals have
such experiences, not all are necessarily qualified as expert witnesses on the
nature of their experience. Thus Wundt embraced the method of intro-
spection-a method whereby one attends carefully to one's own sensations
102
and reports them as objectively as possible. Such objectivity here means
that one describes the sensations felt, rather than the stimulus giving rise
to them; and that one reports thoughts (or images) without reference to
their meaning or context of presentation. Wundt's program hinged on the
possibility of introspecting in this fashion.
Just as it was important to separate psychology from physics, it was
also important to effect its divorce from physiology. Physiology has as its
special province the study of mental processes that are not conscious-not
susceptible to introspective examination-while psychology appropriates
those higher mental processes that could be subjected to personal examina-
tion. Physiology explains how we have sensations, but psychology covers
the description and analysis of sensory experience. Recognizing that some
aspects of human experience prove less hospitable to examination in terms
of introspection, Wundt made a further discrimination within psychology
between those individual experiences that are susceptible to introspection
and those aspects of human experience that are by their nature social or
communal. In Wundt's division of labor, it fell to ethnic or folk psycholo-
gists to study these complex and large-scale human activities, such as
customs, rituals, and, perhaps, certain features of language and thought.
Ultimately, he contributed ten volumes to this effort at constructing folk
psychology.
As George Miller has put it, Wundt looked at the agenda of British
empirical psychology with an eye of a man trained in the traditions of
German physiology (Johnson-Laird and Wason 1977, p. 2). Like Helmholtz
and Donders, Wundt devised simple tasks-but asked subjects to intro-
spect about them during the course of their participation. In one sample
task, subjects were asked to press a button whenever they heard a tone.
On some trials, the subjects' attention was directed to the tone; while on
others, their attention was directed to the movement of their fingers on the
button. It turned out that reaction times were longer when a subject had to
attend to the sound, shorter when one attended to the movement. Wundt
inferred that this greater time was due to apperception-the process of mak-
ing one's experience clear in one's consciousness. Apparently, when one
is concentrating on a stimulus, the stimulus first has to be perceived and
then apperceived, or consciously interpreted, in the light of the response
associated with it. Even further complications resulted when several differ-
ent stimuli were presented, only one of which was targeted for response.
This process took even longer, because, according to Wundt, "cognition"
has to occur: the stimulus has not only to be perceived and apperceived
but also differentiated from other stimuli that are not to elicit responses
(Fancher 1979, p. 139).
Wundt revealed his ties to the British philosophical tradition in the
way in which he made sense of such findings. He came to think of experi-
103
II I THE CoGNITIVE SciENcEs : HISTORICAL PERSPECTIVE
ence as composed of simple basic elements-raw sensory content devoid

of any meaning; and all conscious thoughts were accordingly assumed to
be combinations of these sensations which can be analyzed in terms of
quality, mode, duration, intensity, and the like. Elements can be fused into
complexes in any number of ways, so that one element arouses several
others. In the experiment just described, the subject would reveal through
introspection his awareness of the impact of tones, the feelings of pressing
the button, and any connections perceived between these sensations. The
various images passing through the subject's consciousness would be cru-
cial contributors to the subsequent analysis. Wundt' s interest lay in identi-
fying the various laws of association occurring during an experimental
subject's experience; he sought to understand how the laws came to unfold
while individuals were generating ideas, and how they attained different
degrees of clarity.
It is important to stress that Wundt was neither a simple-minded
atomist nor a person ignorant of the effects of context or cultural influ-
ences. Indeed, in many studies, he paid attention to molar issues and
top-down effects. In his writings on language, for example, he anticipated
some of the ideas of modern psycholinguistics-such as the role of prior
intentions in governing speech output and the effects of syntactic structure
on the production of messages. Still, a scientist must be judged by the
major thrust of his work; and, in the aggregate, Wundt's psychology
emerges as a kind of mental chemistry, focused principally on the discov-
ery of the pure elements of thought, through whose combination com-
plexes of mental activity come to be formed.
This, then, in outline, was Wundt's program. I must stress that this
program was not simply enunciated once and then forgotten. Rather,
Wundt worked tirelessly for fifty years to promulgate it, to train individu-
als in its practice, and to make sure that there existed the appropriate
laboratories, handbooks, journals, and conferences through which to put
it into practice. Wundt trained many, if not most, of the leading researchers
around the turn of the century. He set a high standard for the conduct of
experiments, insisting on careful selection and training of subjects, on the
administration of relevant controls, and on the replication of experiments
to make sure that findings were not accidental. He himself systematized
findings in encyclopedias, handbooks, and textbooks, revising them with
awe-inspiring regularity. Indeed, his literary output in psychology stretch-
ing over a period of sixty-three years has never been equaled.
Innovative Methods: Hermann Ebbinghaus

Everyone marveled at Wundt's scholarly productivity, but not all
embraced his view of psychology. Some of his colleagues-for example,
104
Hermann Ebbinghaus (1913)-simply went their own way. Inspired with
the fertile idea of ferreting out principles of memory through the use of
materials that could not be contaminated by earlier experiences and as-
sociations, Ebbinghaus generated over two thousand nonsense syllables
and measured his own skill at learning sets of them. Practicing them either
at one sitting or over a series of sessions, Ebbinghaus probed the effects
of drill time between presentations, backward and forward interference,
and other "independent variables." His stimuli were no more meaningful
than Wundt's; but, by breaking away from introspection as the sole source
of information, and by instructing novel methods of statistical analysis,
Ebbinghaus succeeded in reorienting psychology in a very productive
manner. It was one's actual skill at a task-not one's introspections about
what one senses or feels-that became the object of study and measure.
Ebbinghaus's methods have exerted far more of an impact on experimental
psychology than has Wundt's introspective tack. Yet, influential as this
work has been, the question must be raised whether the immaculate,
content- and context-free approach of Ebbinghaus ultimately accom-
plished more harm than good; for we now know that methods used to
remember ordinary events differ in important particulars from meth-
ods used to remember meaningless arrangements of letters or organized
material.
The Early Twentieth Century
The Attack on the Wundtians

For a while Wilhelm Wundt and his followers held sway over the
growing band of psychological researchers. But within two decades, the
program of systematic introspection underwent severe attacks from some
one-time students of the master of Leipzig. Around 1900, there began in
Germany a modern-day incarnation of the debate between Locke-who
believed in abstract ideas, devoid of imagery-and Berkeley who, skeptical
of the existence of abstract ideas, believed only in the utility of specific
imagery. The renegade, Lockean view was defended by a group of psy-
chologists housed in the small city of Wiirzburg, and ultimately known to
historians of psychology as the "Wiirzburg school." Headed by Oswald
Killpe, a psychologist influenced by Franz Brentano, this school became
embroiled with Wundt and his devoted American follower Edward Brad-
ford Titchener in a notorious controversy over "imageless thought." In its
specific form, this controversy simply re-created a classic empiricist debate.
105
II I THE CoGNITIVE SciENCES : HisTORICAL PERSPECTIVE
But in its more general form, Wiirzburg had launched a severe critique of
the entire manner in which Leipzig went about its investigations and drew
its conclusions.
The story begins innocently enough in 1901, when the young psy-
chologist Karl Marbe asked his subjects to compare weights with one
another-a familiar procedure, dating back to Fechner's first reveries.
Marbe found that, in judging weights, subjects reported no imaginal con-
cepts as the basis of judgment: in other words, no images flitted through
their introspecting minds. Instead, counter to the Wundtian expectation,
subjects reported that various vague attitudes passed through conscious-
ness-attitudes like hesitation, doubt, waiting for an answer, feeling that
the answer had arrived. Marbe was forced to conclude, "The present data
are quite sufficient to draw the conclusion that no psychological conditions
of judgments exist .... Even ... the observers concerned ... were extremely
surprised to note the paucity of experiences that were connected with the
judgmental process" (Mandler and Mandler 1964, p. 143).
That some subjects failed to report imagery while rendering a judg-
ment may hardly seem like headline news in the scientific community, but
it ran against the widespread dictum that all thought features imagery
accessible to consciousness. Then, in the following years, a whole raft of
studies emanating from Wiirzburg repeated and elaborated upon this ini-
tial negative report (see Boring 1950; Humphrey 1951; Mandler and Man-
dler 1964). There was Henry Watt who reported that the conscious task
(or Au/gabe) that a subject was posed had an important effect on the kinds
of associations one made. There was Narziss Ach who reported the opera-
tion of a determining tendency: the task had the effect of orchestrating various
associations and skills into a purposeful orderly sequence-a kind of "di-
recting" will-which led smoothly to its final execution. There was August
Messer who saw consciousness as the visible portion of an iceberg, with
most thought processes occurring beneath the surface, and with conscious
processes themselves exhibiting varying degrees of clarity. And there was
Karl Buhler, one of the seminal figures of psychology, who dismissed the
simple problems posed by the Leipzigers and posed truly complex tasks to
his subjects-for example, the discussion of philosophical problems. Like
his Wiirzburg colleagues, Buhler uncovered peculiar varieties of conscious-
ness like doubt, astonishment, and even consciousness of consciousness.
Oswald Kiilpe sought to put these different demonstrations together.
In a critique of the Wundt-Leipzig program, he pointed out that it was not
enough merely to pose problems to individuals and let them introspect as
they wished. It was necessary to pose problems making different kinds of
demand and to monitor their varying effects. One could not simply assume
that all important aspects of mental processes are conscious, that images
106
always guide thinking, and that mental contents (like sensations) are nec-
essarily the constituent elements of thought. There was a more positive
aspect to the critique as well. Thanks to the work of Kiilpe and his associ-
ates, mental acts like attending, recognizing, willing, comparing, and differ-
entiating came to constitute a proper sphere of psychology. While, to be
sure, these acts lack vivid perceptual characteristics of sensations, images,
and feelings, they are no less important on that account. Most generally,
there was an increased willingness to recognize top-down "structuring
tendencies" in the solution of these problems and to use problems that
truly engaged a subject's ratiocinative capacities. Here is a harbinger of the
molar concerns of many contemporary cognitive scientists.
The Wundtians did not accept such criticism without a rejoinder.
Points were scored on both sides-Wundt's more methodological, Wiirz-
burg's (to my mind) more substantive; but in the end, this disagreement
had a radical consequence. It called into question the merits of any psychol-
ogy that relies heavily on introspection, and particularly on the introspec-
tions of trained subjects, an uncomfortably high proportion of whom are
drawn from the ranks of the experimenters themselves. As vast an enter-
prise as a new science could not countenance so vague and subjective a set
of basic procedures.
Functionalism: William fames

The programs of Wundt and the Leipzig school at first had a signifi-
cant impact on the other side of the Atlantic. Most young American schol-
ars interested in psychology served an apprenticeship in Wundt's labora-
tory; and some, like the transplanted Englishman Edward Titchener,
became lifelong adherents to the introspectionist program. But the dean of
American psychologists of the period, William James, soon became disen-
chanted with Wundt. He saw Wundt as a classic German professor, who
had to have an opinion on every issue, but who, in his dogged pursuit of
the elements of thought, missed many of the most exciting and important
issues in psychology. In his landmark textbook, James wrote, with ill-
conceived contempt, that Wundt's experimental psychology" could hardly
have arisen in a country whose natives could be bored . ... There is little
of the grand style about these new prism, pendulum, and chronograph-
philosophers. They mean business not chivalry" (James 1890, vol. I,
p. 193).
There was plenty of style in William James's writings. Indeed, jux-
taposed to the involuted sentences and turgid experimental details of the
Leipzig-Wiirzburg schools, his words are like a gust of fresh air. James took
his methods where he found them and had little interest in starting a
107
program or launching an institution. Instead, he was fascinated by the

questions of psychology, as encountered in daily life, and sought illumina-
tion from literature and history as well as from experimental science. He
allowed his lively imagination to frolic over the full playing field of psy-
chological phenomena-from brain processes to the will, from the sensory
organs to the study of attention, from an investigation of habit to the
description of the momentary stream of consciousness. In his Principles of
Psychology, he took a characteristically American pragmatic approach,
which he himself was to propound in philosophical writings. He suggested
that psychological mechanisms exist because they are useful and help
individuals to survive and carry out important activities of living. As he
declared, "Our various ways of feeling and thinking have grown to be
what they are because of their utility in shaping our reactions to the outer
world." Without hesitation, he assumed purpose to be the mark of mentality
-Romeo wants Juliet as filings want a magnet, he asserted. "The pursu-
ance of future ends and the choice of means for their attainment are thus
the mark and criterion of the presence of mentality in a phenomenon"
(quoted in Herrnstein and Boring 1965, p. 610).
The mood permeating James's writings may well have had more of an
effect on subsequent generations of American psychologists than did any
of his particular claims. For he was signaling an impatience with the whole
tone of German psychology, with its inconclusive introspection. Instead of
trying to figure out the contents of mental life, and how they were struc-
tured, James directed attention to the various functions carried out by
mental activity-to the active exploring nature of mind in the positing of
problems and the attainment of goals, and to the perennial dialectic be-
tween means and ends. James stood at the vanguard of a new American
movement which opposed Wundt-Titchener structuralism-a movement
that came to be known, appropriately enough, as "functionalism."* As
against the structuralist attempt to ferret out the elementary components
of experience, the functionalist psychologist sought to investigate the op-
erations of mental activities under actual life conditions.
According to functionalist precepts, it is important to focus on func-
tions or dispositions, like perceiving or recalling or thinking. A function
persists and recurs, while the contents of consciousness and specific re-
sponse patterns occur but once. Science ought to be based on these more
enduring and meaningful series of actions.
Functionalism undeniably provided a needed shift in emphasis, and
a counterweight to the search for ultimate units of consciousness. But just
a few short years after the canons of functionalism had been put forth by
"Not to be conflated with the philosophical brand of functionalism described in
chapter 4.
108
William James (and also by James Angell and John Dewey), a far more
radical shift was to occur in American psychology. This shift to behavio-
rism may initially have been necessary in order to put firmly to rest the
excesses of structuralism and introspection, whether in their Leipzig or
their Wiirzburg guise. Yet, from the point of view of a history of cognitive
science, it is difficult to think of this phase as other than primarily negative
and regressive.
The Behaviorist Revolution

In 1913, John B. Watson, scarcely out of graduate school but already
a force in American psychology, launched the behaviorist revolution. Wat-
son asserted that the proper subject for psychology was not the operation
of the mind but rather the examination of objective, observable behavior.
Building on the physiological studies of the operation of the reflex arc,
Watson proposed that all concerns of psychology could be explained
through understanding the reflexes occurring in the upper portions of the
nervous system. This was molecular and bottom-up psychology, pure and
simple. As he declared:
Psychology as the behaviorist views it is a purely objective natural science.

Its theoretical goal is the prediction and control of behavior. Introspection forms
no essential part of its method nor is the scientific value of its data dependent upon
the readiness with which they lend themselves to interpretation in terms of con-
sciousness. The behaviorist, in his efforts to get a unitary scheme of animal re-
sponse, recognizes no dividing line between man and brute. The behavior of man,
with all its refinement and complexity, forms only a part of the behaviorist's total
scheme of investigation. (Quoted in Fancher 1979, p. 319)
Watson's dramatic call had far-reaching consequences. He was reject-

ing much of the program and nearly all of the method of traditional
psychology. No more felt sensations or intentions: henceforth, only obser-
vation of overt behavior was relevant. Description and explanation of states
and contents of consciousness were to be replaced by the prediction, and
eventually the control, of behavior. Psychology would no longer care
about what is putatively on a person's mind because all mentalistic terms
themselves were hereby expelled from the psychologist's vocabulary.
It would be difficult to overestimate the extent to which Watson's
program came to dominate American psychology and even to exert influ-
ence abroad. The old psychology might have collapsed of its own weight,
but this disintegration was certainly abetted by Watson's rhetorical vigor
and by the efficacy of his demonstrations. On his practically oriented
account, it was possible to condition organisms, including humans, to do
109
almost anything one wanted-for example, to fear an object-just by

arranging the environment in a certain way. Indeed, he did just this in his
famous study of baby Albert, whom he conditioned to fear a white rat
(Watson and Rayner 1920).
Such dramatic demonstrations of what psychology could do appeared
even more convincing when set alongside the inconclusive debates of the
German school. A whole generation of scientists-the leading psycholo-
gists of the next generation-were trained in the orbit of Watson; and
investigators like Clark Hull, B. F. Skinner, Kenneth Spence, and E. L.
Thorndike helped to ensure that the psychology of America between 1920
and 1950 was irremedially behaviorist. Child care, treatment of prisoners,
teaching of children, and many other societal activities came to be domi-
nated by behaviorist rhetoric and behaviorist practices. No less an author-
ity than the New York Times declared in 1942 that behaviorism marked "a
new epoch in the intellectual history of man" (Fancher 1979, p. 322).
Even members of the American psychological community who had
fundamental doubts about Watson's program-scholars like Edward C.
Tolman at Berkeley and Watson's one-time student Karl Lashley at Har-
vard-worked primarily with animals and adhered rigidly to the experi-
mental precepts of the behaviorists. To be sure, Lashley ultimately voiced
his qualms: at the Hixon Symposium, he called attention to behavior that
could not be explained in terms of stimulus-and-response links. For his
part, Tolman (1932) found it impossible to explain the ability of animals
to find a reward simply through invoking a memorized sequence of actions.
Inasmuch as the animals seemed to know where to go, even when the
orientation or the cues of a maze had been altered, Tolman found it neces-
sary to posit "cognitive maps" which guide the behavior of animals en-
gaged in problem solving. Moreover, in describing the activities of animals,
he found it necessary to invoke an intentionalistic vocabulary, using terms
like ''expectancies," "purpose," and "meanings." Correlatively, he took
the then-daring step of introducing "intervening variables" between stim-
ulus and response. And yet, despite such heterodoxies, these influential
scholars remained respectful of the behaviorist canons and, in fact, con-
tinued to speak of themselves as behaviorists, if of a somewhat renegade
stripe.
As I noted in chapter 2, far too much of consequence in human
behavior was denied by the behaviorist approach. By the middle 1950s, its
program had begun to come apart; and today the theoretical claims of
behaviorism (though not its various applied achievements) are largely of
historical interest. The cognitive revolution spawned by psychologists like
Miller and Bruner has carried the day, just as surely as the behaviorist
revolution did half a century ago. Indeed, somewhat paradoxically, cog-
110
nitivism has prevailed for some of the reasons behind behaviorism's origi-
nal success: an exciting and refreshing new approach was seen as supplant-
ing a time-worn tack mired in questions of little interest to anyone outside
the field. But despite behaviorism's relatively rapid decline, only a few
lines of work in psychology fall largely outside the behaviorist camp.
Among these, I shall focus on the most direct link between the cognitively
oriented psychology of 1900 and that of today-the school of Gestalt
psychology.
Gestalt Psychology: A View from Above
Origins
Gestalt psychology can be introduced as a school rooted in several
dramatic demonstrations, which make a vivid point and thereby inspire a
certain way of thinking about mental phenomena. The initial demonstra-
tion of Gestalt phenomena was made in 1890 by Christoph von Ehrenfels,
an Austrian student of Brentano's. Ehrenfels's particular interest was the
perception of melody. He argued that the perceptual"form quality" in-
volved in a melody cannot be properly viewed as simply the sum of its
several tonal elements: indeed, it is an overall quality, a Gestalt, that
transcends its particular elements. As he pointed out, one could take the
same set of elements, or tones, and produce an entirely different melody.
Conversely, one could select an entirely different set of tones-for in-
stance, those in another key-and produce a figure that would be ap-
prehended as the "same melody" as the original one.
Ehrenfels is generally regarded as a predecessor, rather than the
actual founder, of the Gestalt movement. That honor belongs to Max
Wertheimer, who in 1912 published a paper on the visual perception of
movement. Working with two young assistants, Wolfgang Kohler and
Kurt Koffka, Wertheimer carried out a set of studies on apparent motion,
or the "Phi" phenomenon-roughly, the perceptual experience of move-
ment which arises when a set of lights or forms appears one right after
another (as in neon billboards or, for that matter, motion pictures). Wer-
theimer did not discover apparent motion-the phenomenon had been
recognized for some time; but he showed that the standard psychological
account of this phenomenon was untenable.
In particular, Wertheimer ruled out the common hypothesis of the
perception of movement being due to eye movements: he showed that
111
I I I THE COGNITIVE SCIENCES : HISTORICAL PERSPECTIVE
movement is perceived even if the interval is too brief to permit a move-

ment and even if the subject maintains a rigid fixation. Wertheimer also
made a convincing case that perception of movement is not a sum or an
association of different elementary sensations: his subjects-who were
simply asked to report, and not to analyze, what they saw-all agreed that
they apprehended a movement directly, rather than its being a movement
of something, or of something from one place to another. This finding
suggested the need for a new way of thinking about such perceptual
phenomena.
Wertheimer and his colleagues attributed perceptual experiences like
apparent motion to the way in which the brain organizes perceptual input.
In their view, a kind of short circuit occurs in the "physiological fields" of
the brain. Hence, there was no need to posit a construction from single
elements: the patterns of excitation in the brain ensure that movement can
be perceived directly.
The Gestalt psychologists examined a whole raft of "form qualities,"
whose phenomenal appearance could be explained in terms of analogous
brain processes, and put forth laws purporting to explain how perception
is organized. For instance, they showed that objects that are close together
tend to be grouped together (the law of proximity); the more symmetrical
a closed region, the more it tends to be seen as a figure (the law of symme-
try); and the arrangement of figure and ground seen is the one featuring
the fewest changes or interruption in straight or smoothly curved lines (the
law of good continuation) (Hochberg 1978). Though usually referring
initially to visual demonstrations, versions of these laws also applied to
auditory sequences-for example, rhythmic patterns. Pervading their
powerful demonstration was the Gestalt psychologists' doctrinaire opposi-
tion to atomistic, bottom-up, or purely molecular analysis; they favored a
view of perceptual organization in which the way the parts are seen is
determined by the configuration of the whole, rather than vice versa.
Kohler 5 Comprehensive Researches

Perhaps because Gestalt laws could be applied readily to the range of
perceptual phenomena with which psychologists were already familiar-
such as various illusions and constancies-the Gestalt movement rapidly
gained in popularity and influence, particularly in Continental Europe. But
the Gestalt psychologists were not content to rest on their perceptual
laurels. When Wolfgang Kohler found himself stranded on Tenerife (off
North Africa) during the First World War, he undertook a landmark set
of investigations with chimpanzees. Kohler's interest centered on the way
these apes tackled problems whose solution required some changing or
112
"restructuring" of the elements in a situation. Thus, in order to fetch a
banana that was out of reach, an ape would have to move a chair from one
site to another, join two sticks together, or engage in some other reorgani-
zation of the situation that it beheld. Kohler (1925) found standard trial-
and-error accounts inadequate to account for the apes' behavior. Instead,
what seemed to happen was far more of a humanlike thought process,
where the ape would stop, reflect, and then, as if struck with a sudden flash
of insight, reach for the chair or the rope that provided the solution.
According to Kohler's analysis, the chimpanzee "restructured" the field
that had been presented to it; and did so courtesy of a moment of insight
-what the Wiirzburg school had termed an "A-ha experience."
Kohler also distinguished between bright and stupid apes. While a
bright ape would regularly exhibit these moments of insight, the dull ape
worked in quite a different fashion. Even when it saw the correct behavior
modeled, the ape would not be able to gain the treasure. Instead, in piece-
meal fashion, the ape would imitate the component actions without ever
apparently appreciating how to link them in order to secure a reward.
According to Gestalt psychology, the most primitive forms of learning
can be explained in terms of mere repetition or piecemeal associations. In
contrast, what characterizes higher learning or "intelligent" processes,
wherever found, is the capacity to grasp the basic fundamental relations
in a situation. These are the criteria of insight: "the appearance of a com-
plete solution with reference to the whole layout of the field" (Mandler
and Mandler 1964, p. 248).
Other scholars in the Gestalt tradition extended this line of study to
problem solving in human beings. Wertheimer (1945) himself examined
the solution of geometry problems, arithmetic puzzles, and even the steps
through which Einstein allegedly passed in arriving at his theory of relativ-
ity. Karl Duncker (1945) focused on the solution of engineering problems
-for example, the way in which one can direct X rays to kill a tumor
without destroying an unnecessary portion of intervening tissues.
Abraham Luchins (1942) examined the phenomenon of functional fixed-
ness: the way that the customary uses to which a material are put can
inhibit an individual (ape or professor) from perceiving how to use that
same implement in a novel way in order to solve a problem.
Note that the kinds of molar problem posed stood in sharp contrast
to those molecular tasks typically posed by the early German structuralists
and the American behaviorists. These molar problems are somewhat closer
to those posed by the Wiirzburgers but unencumbered by the weight of
introspections and elaborate interpretive terminology, and are forerunners
of problems favored today by many researchers in artificial intelligence.
The modes of explanation are also of a different order. Wertheimer and the
113
I I I THE CoGNITIVE SciENcEs : HISTORICAL PERSPECTIVE
other Gestaltists proposed a far more active, strategic approach to prob-

lems, in terms of recognition of structural characteristics: how these struc-
tures are initially realized; which gaps are sensed; which elements can be
fitted with one another or rearranged; which dynamic operations can then
be brought to bear to fill the gaps, leading from "an unclear inadequate
relation to a clear transparent, direct confrontation-straight from the
heart of the thinker to the heart of his object, of his problem," as
Wertheimer rather grandly put it (1945, p. 191).
The Gestaltist saw the best thinking as productive and novel rather
than as reproductive. There are regular perceptual reorganizations as one
confronts a problem; and sooner or later, a pivotal reorganization is accom-
panied by insight as elements critical to a solution come to the fore. There
was a Kantian undergirding to the belief that the mind is so constructed
that certain logical relations are imposed upon the world, rather than being
"read off" of one's experience of the world.
In the view of most contemporary observers, the particular theoretical
program of Gestalt psychology was not well founded. Principles of percep-
tion such as proximity and symmetry do provide a rough-and-ready guide to
how information is organized, but there are too many exceptions or in-
determinate cases; speculations about the operation of the brain, and its
effects upon phenomenal perception, have been undermined by neurologi-
cal findings; the major explanatory concepts are too vague to be operation-
alized. As psychologist Ulric Neisser concludes, in a sympathetic review of
Gestalt psychology:
The gestalt concept of organization plays almost no role in cognitive psychol-

ogy today. It ran aground on the emerging facts of neuropsychology (the nervous
system turned out to be more elementistic than Kohler believed) and it was over-
taken by mechanistic computer models that gave plausible alternative accounts of
the same facts .... The currently popular explanations of motion phenomena in
terms of eye movements and specific motion detectors are about as far from the
spirit of Gestalt psychology as one could get. (1980, p. 4)
Yet Neisser concludes with a more positive assessment: "Whatever our

differences, all of us are still working on the problem Wertheimer set: to
see the world 'from above,' as Kohler put it, instead of from below. Be-
cause, indeed, the view is better from there" (p. 6). Here is a suggestion
that the top-down approach may have time on its side.
Frederic Bartlett s Schematic Approach

As the Gestaltists were sustaining a concern with large-scale prob-
lems, holistic methods of solution, and the constructive aspects of think-
114
ing, a solitary psychologist working in Britain was also keeping the cogni-
tive faith. In his own explorations of memory, Frederic Bartlett had sought
to use, but had found inadequate, the strict experimental methods pio-
neered by Ebbinghaus. They seemed to miss much of what was central in
remembering meaningful content, and Bartlett came to the conclusion that
it is not possible to employ totally arbitrary materials and still capture the
salient features of memory on the wing. A fundamentally different ap-
proach was wanted.
Bartlett had vague intimations that memory was more of a social or
cultural phenomenon, and that the "set" accompanying a stimulating ex-
perience has a crucial effect on what one remembers and how well one
remembers it. Yet, according to his own recollections, Bartlett was stymied
for an experimental approach until he had a conversation with his friend
Norbert Wiener. In an important moment in the intellectual history of
cognitive psychology, Wiener gave Bartlett a crucial idea, as the latter
recalls:
I was fascinated by the variety of interpretations which different people then

achieved .... One day, when I had been talking about my experiments, and the
use I was making of sequences in a study of conventionalization regarded as a
process more or less continuous in time, [Wiener] said: "Couldn't you do some-
thing with 'Russian Scandal'* as we used to call it?" That was what led to the
method which I later called "The Method of Serial Reproduction," one which, in
varied form, was to contribute much to the final working out of my experiments.
{1958, pp. 139, 144)
In the set of studies that Bartlett eventually launched, subjects heard

exotic stories and then were asked to recall them at various subsequent
intervals. (In variations of the procedure, they were shown geometrical
figures or exposed to logical arguments and asked to retain them for later
"debriefing.") Bartlett (1932) found that individuals were not able to recall
such input accurately and, more revealing, that the inaccuracies exhibited
systematic patterns. Take, for instance, "The War of the Ghosts," an
Indian folk tale which (to a modem Western ear) seems to be filled with
strange gaps and bizarre causal sequences. Subjects would regularly supply
their own causal links, drop difficult-to-assimilate information, and revise
the plot until it had come to resemble that of a standard Western tale.
Moreover, on subsequent recall, these alterations would come increasingly
to the fore until the story reached a relatively stable form--one far closer
to a prototypical Western story than to the folk tale originally presented.
Drawing on a term that had been used (in a somewhat different
*"Russian Scandal" is a parlor game in which players pass a message from one to
another and then observe how it has been transformed.
115
sense) by Kant and introduced into the psychology of the period by the
British neurologist Henry Head, Bartlett tried to make sense of his
findings in terms of the notion of schemas (sometimes called "schemata").
Bartlett claimed that the typical memory system used by humans in-
volves the formation of abstract cognitive structures, or schemas. These
schemas arise from prior encounters with the environment, as a result of
which certain kinds of information have come to be organized in spe-
cified ways. Thus, in listening to "The War of the Ghosts," subjects
would draw on their schemas for dealing with daily experience, in gen-
eral, and for dealing with adventure or ghost stories, in particular. To
make sense of this story of a war party proceeding on a canoe in order to
kill people, the listeners would use experiences of their own-say, a
canoe trip at night on a river-as well as structured information from
earlier stories-a typical adventure story involving primitive folks and
ghosts. To the extent that the information in Bartlett's story was consist-
ent with these previously constructed schemas, recall would be aided and
might tum out quite accurate. On the other hand, divergences between
the prior schemas and the details of the present story would cause sys-
tematic distortions in the initial recollection of the story and probably
introduce even further deviations in subsequent retellings.
Speaking more generally (and no doubt with the Ebbinghaus non-
sense-syllable tradition in mind), Bartlett declared:
Remembering is not the re-excitation of innumerable fixed, lifeless, and frag-

mentary traces. It is an imaginative reconstruction, or construction, built out of the
relation of our attitude towards a whole active mass of past experience .... It is
thus hardly ever really exact, even the most rudimentary cases of rote recapitula-
tion, and it is not at all important that it should be so. The attitude is literally an
effect of the organism's capacity to turn round up upon its own "schemata" and
is directly a function of consciousness. (1932, p. 213)
Bartlett did more than keep alive a model for molar studies; he directly
anticipated the self-reflective system that cognitive scientists like George
Miller view as central to human cognition.
jean Piaget's Developmental Concerns

Jean Piaget got his start in psychology in a most unusual way. Trained
as a biologist, with a particular interest in mollusks, he had taken a job as
a tester in the laboratory of Theodore Simon, a former colleague of Alfred
Binet, who had invented the IQ test. Never one to accept a task as given,
Piaget became interested in the kinds of errors children make on intelligence
test items. Since Piaget's lifetime goal was to found an epistemology on
116
biological principles, he decided to take a brief detour to study the devel-
opment of thought in children. As he liked to say, this detour lasted a
lifetime.
In the course of a sixty-year career, rivaling Wundt's in productivity
and exceeding Wundt's in its influence upon subsequent research, Piaget
provided portraits of the developmental course of children's thought, in a
variety of domains (see Piaget 1970 for an overview). Nearly always, his
work exhibited two hallmarks. On the one hand, conforming closely to my
definition of a cognitive scientist, Piaget took as his research agenda the
great issues of Western epistemology: the nature of time, space, causality,
number, morality, and other Kantian categories. He regarded these catego-
ries not as givens in the mind but rather as categories to be constructed,
Helmholtz style, over the course of a child's development. On the other
hand, Piaget insisted on painstakingly careful observations of children:
sometimes as they were engaged in free exploration and play; more often
as they were involved in experimental tasks he had cleverly contrived.
It is probably as an inventor of brilliant experimental paradigms and
riveting demonstrations that Piaget will be best remembered by both the
public and the psychological community. His demonstrations have had
profound effects on work in cognitive developmental psychology. Who,
before Piaget, suspected that infants believe an object remains in its origi-
nallocation even when it has been moved before their eyes to a new one?
Who thought that toddlers cannot appreciate how a collection of objects
looks from a perspective different from their own? Who anticipated kin-
dergarteners' beliefs that the amount of liquid changes when it has been
poured into a vessel of different shape? Even when Piaget's particular
demonstrations have not always stood up in just the way he described
them, further knowledge has invariably been built upon his pioneering
discoveries.
But in Piaget's mind, he was an epistemologist or, as he preferred to
say, a genetic epistemologist-not a child psychologist. He sought to un-
ravel the basic laws of thought: informal tasks with young children were
simply his preferred means of securing data about the nature of knowl-
edge. His principal contribution to psychology, as he saw it, was an un-
raveling of the basic structures of thought that characterize children at
different ages or stages of development, and his suggestion of the mech-
anisms that enable a child to effect a transition to higher stages of devel-
opment-from the sensorimotor stage of infancy to the intuitive stage of
early childhood, or from the concrete operational stage of middle child-
hood to the formal operational stage of adolescence. In positing specific
stages, Piaget relied heavily on logical formalisms; in expounding the
rules of transformation from one stage to another, he relied on biologi-
117
cal mechanisms, of the sort thought to govern change in any organismic

system.
Piaget's grandiose claims have proved less robust than his specific
experimental demonstrations. The logical formalisms underlying specific
stages are invalid, the stages themselves are under attack, and his descrip-
tions of the biological processes of stage transformation have eluded even
sympathetic scholars (see Brainerd 1978). Even his impressive program of
genetic epistemology has fallen into disuse-apparently too daunting a
challenge for most investigators to pick up. But Piaget did more than
simply help to keep alive the cognitive flame during the behaviorist
hegemony. More than nearly any of his predecessors, he launched an
entire field of psychology-that concerned with human cognitive develop-
ment-and provided the research agenda that keeps it occupied until this
day. Even disproofs of his specific claims are a tribute to his general
influence.
Thanks to research programs like Piaget's, to lines of study like Bart-
lett's, and to molar concepts like schemas, operations, and strategies, the
concerns of the Gestalt psychologists and of their Wiirzburg predecessors
remained alive in the English-speaking world during the height of the
behaviorist era. Among the few other similarly minded scholars were, as
I have noted, the comparative psychologists Karl Lashley and Edward
Tolman, along with the perceptual psychologist].]. Gibson, who avoided
the vocabulary of stimulus-response psychology and directed his attention
instead to the rich lodes of information present in the visual environment
(see chapter 10 and references contained therein); and the English social
psychologist William McDougall (1961), who insisted on the place of
purposeful, goal-striving activity and complex motivations in an otherwise
antiteleological environment. But these psychologists, however illustrious,
still comprised a small and isolated group during the behaviorist era of the
1930s and 1940s.
The Turn to Cognition
Inspiration from Computers

Defying the warnings of Kant, the first century of psychological re-
search-roughly from 1850 to 1950-demonstrated beyond question that
psychological studies were possible. Mental processes could be investi-
gated in the laboratory, many of them could even be timed, and the role
118
of the knowing subject could be adequately controlled. This new psychol-
ogy took many forms in many places, and there were definite swings of
the pendulum between molar and molecular approaches, between an em-
phasis on the determining role of the environment and one on the contri-
butions of the subject to the task at hand. Excesses of introspectionism at
the tum of the century were replaced in tum by excesses of behaviorism
in the early part of the twentieth century. In particular, mentalistic con-
structs were disallowed, and aspects of language and problem solving were
either omitted altogether or treated in much attenuated form.
By the late 1940s, at the time of the Hixon Symposium, it was becom-
ing clear that neither the physiological nor the psychological forms of
behaviorism were viable. Available as alternative models were Gestalt
psychology, as well as the still-isolated efforts to study higher forms of
problem solving by Bartlett, Piaget, and a few other investigators. But it
took the advent of computers (which could themselves exhibit problem-
solving behavior), and the rise of information theory (which provided an
objective basis on which to stipulate components of language or concepts)
to grant legitimacy to cognitive studies. As Ulric Neisser comments, in a
brief sketch of the history of cognitive psychology:
It was because the activities of the computer itself seemed in some ways akin
to cognitive processes. Computers accept information, manipulate symbols, store
items in "memory" and retrieve them again, classify inputs, recognize patterns and
so on. . . . Indeed the assumptions that underlie most contemporary work on
information processing are surprisingly like those of nineteenth century introspec-
tive psychology, though without introspection itself. (1976, pp. 5, 7)
As I have already noted, the rise of cognitive psychology in the middle

1950s was a complex matter, reflecting changes in the Zeitgeist, new meth-
ods in allied sciences, and the enhanced legitimacy of concepts like inten-
tion, purpose, goal, and problem solving now that "mere" mechanical
gadgets could lay claim to these processes. The particular achievements of
innovative psychologists like Donald Broadbent, Jerome Bruner, and
George Miller doubtless promoted the tum to cognitive psychology, as did
the excitement surrounding the rediscovery of Piaget, Bartlett, and certain
works of the Gestalt psychologists. In some cases, high-power informa-
tional and mathematical techniques were simply transferred to elementary
aspects of learning and problem solving left over from an earlier day; but,
at its best, the new technology was solidly concentrated on problems of
substance.
Further Stages of Information Processing By and large, the first efforts on both
sides of the Atlantic sought to ascertain the details of information-process-
ing. The paradigm pioneered by Donald Broadbent proved an especially
119
II 1 THE CoGNITIVE SciENCES : HisTORICAL PERSPECTIVE
popular source for experimentation; and soon experimentalists were

finding difficulties of various sorts in his imaginative line of research.
Neville Moray (1969} showed that one does obtain information from an
unattended message prefaced by one's own name. Anne Treisman (1960,
1964} showed that when identical messages are presented to both ears
without the subject being warned, the subject does indeed detect this ploy
and so can be said to be detecting informative aspects of the message
binaurally after all. This finding led Treisman to argue that the filter serves
not as a simple "all or none" switch but rather as a means of attenuating
irrelevant or unattended signals: the unattended signal just receives less
processing than does the attended signal. Treisman also showed that sub-
jects can shift from one ear to the other if the content of a message itself
makes this shift.
In his influential textbook published in the middle 1960s, Ulric
Neisser (1967) put forth a much more complex informational account. He
argued that a subject comes to understand a signal by synthesizing (pro-
ducing from scratch) an internal representation that matches that signal.
Thus, attention itself is seen as a process of constructing the information
presented to the listener (p. 213}. Remote or apparently irrelevant streams
of speech are analyzed only by passive "pre-attentive" mechanisms and
are not themselves synthesized in this manner; but once a message has
passed through a pre-attentive scan, the subject concludes that the mes-
sage is worth attending to, and the process of synthesizing commences.
Finally Broadbent {1980) has come to revise his own unidirectional model
of attention, and now sees information as going back and forth in either
direction. He allows individuals a multiplicity of ways of accomplishing
the same task as well as a more active (synthesizing) role in the apprehen-
sion of information.
In the United States, shortly after the rebirth of cognitive psychology,
several very influential experiments were conducted. In 1960, George
Sperling became interested in the question of how much information an
individual can take in visually at one time. Sperling {1960) instituted a
measure called a "partial report": just moments after the initial presenta-
tion of a grid of nine letters (arrayed three by three), subjects were in-
structed (by a tone) on which row of letters to report. Using this procedure,
Sperling showed that individuals are able to recall far more digits than had
been predicted. Under normal conditions of "free report," subjects re-
ported only four or five digits of the possible nine-a finding that suggests
that only about one half of all the information had been taken in; but
under this condition of partial report, the subjects were almost always
completely correct. Sperling inferred that information presented to the eye
is held in a sensory memory where it decays within a second. Provided that
120
the information can be accessed immediately, twice the suspected intake
of information can be documented.
Sperling's clever demonstration helped to kindle interest in the struc-
ture of the early information-processing system. Interest in the fine details
of stages of processing themselves received a large boost from studies a few
years later by Saul Sternberg (1966, 1969). In a typical Sternberg study, a
subject is shown a "memory set" of five numbers. After initial presenta-
tion, one is presented with a single target number and asked to indicate
whether it was in the initial memory set. On succeeding trials, the number
of items to be chosen from is systematically varied, and the effect of the
size of the memory set on speed and accuracy of recall is assessed.
Sternberg reported two principal findings. First of all, he showed that
each additional item of information to be surveyed adds about thirty to
forty milliseconds to the search. Apparently, a subject is engaging in serial
search, making one comparison after another, rather than in simultaneous
or parallel search. Second, a subject engages in exhaustive search. Contrary
to intuition, one does not terminate the search, even after having discov-
ered an identity between the target number and one of the numbers in the
initial set. Thus, search time increases linearly with the size of the set but
does not depend on the position of the target within that set.
Sternberg's demonstrations rekindled the hope that basic human cog-
nitive operations could be timed in the manner suggested by Donders's
work nearly a century earlier. In point of fact, Sternberg's technique repre-
sented a methodological improvement on Donders's in not requiring the
devising of new, possibly incomparable tasks in order to isolate the hypo-
thetical stages. According to Sternberg's analysis, such tasks involve a
series of four stages: encoding; search through memory to find a match;
selecting a response; and executing the response. He then proposed a
model according to which the period between the occurrence of a stimulus
and the production of a final solution is taken up by a series of discrete
operations, or stages, that sum to the total time. The duration of any
particular stage is not affected by the duration of any prior stage.
Using what he called an" additive factors approach," Sternberg sought
to uncover the individual effects and the interaction of various factors that
might influence this task. If two factors had additive (or cumulative) effects,
they would be considered as reflecting different stages of processing; but if
they interacted, then they were seen as affecting the same stage. Through
this analytic method, a single task could be analyzed in its components, each
stage could be thought of as an elementary operation, and a laying out of
each factor would ultimately yield a flow chart for the entire task.
A Model of Memory The Sperling and Sternberg studies were concerned
only with the brief interval following presentation of stimuli (see Posner
121
1969 for another set of influential studies). Other cognitive psychologists

directed their attention to the properties of the memory system over longer
intervals. In the influential model of memory put forth by Richard Atkin-
son and Richard Shiffrin (1968), memory was viewed as having three
"stores." To begin with, there is a store where a stimulus is immediately
registered within the appropriate sensory system: this is the locus of Sperl-
ing's buffer. Then there is a short-term store, a working system in which
entering information normally decays and disappears rapidly, but may
pass into a long-term store. Information decays in the short-term store in
fifteen to thirty seconds, but can be placed in a "rehearsal buffer" and
remain there for a longer time. The longer it is held there, the greater the
probability that it will ultimately be transferred into the long-term store.
Short-term memory is viewed as a buffer consisting of about seven slots,
each of which can hold a single piece or chunk of information. It is as-
sumed that the processing capacity of short-term memory is fixed and can
be devoted only to maintaining a large buffer at the expense of other
activities.
Information entering into the long-term store is envisioned as being
relatively permanent. There are no limits on amount, even though infor-
mation housed there can be rendered inaccessible because of the interfer-
ence of fresh incoming information. Long-term memory consists of several
storage mechanisms-such as one that places information in certain loci,
according to content-as well as several mechanisms of retrieval, including
various searching procedures. Because of its clarity, simplicity, and suscep-
tibility to testing, the Atkinson-Shiffrin model became widely accepted (or
modal) within experimental psychology. (In some respects, it held the same
niche in the 1970s that Broadbent's model had assumed a decade before.)
The authors of a widely used text described the fate of this model:
It was only natural that the modal model became the focus for a generation
of memory textbooks. Textbooks guided by the model characteristically follow the
perceptual input from the sense organs to the brain .... The main problem with
this type of textbook organization is that by the end of the 1970s the "sequential
stage" model [was] modal no longer. (Glass, Holyoak, and Santa 1979, p. iv)
It is difficult to say why the information-processing model described

here eventually waned. After all, it was successful in the sense that re-
searchers could readily devise experiments that yielded quantitative infor-
mation about the number of elements of information processed in a given
amount of time or the consequences of various kinds of manipulation of
stimulus load. Indeed, thanks to the model, most researchers have tended
to think of information processing as a sequential process, where informa-
tion first enters at one site and then is passed on, or shunted off, until it
122
reaches some central processor, where more complex forms of analysis
{including top-down processing) may take place. There has been a correla-
tive tendency to think of an individual's information-processing capacity
as being sharply limited-a certain number of chunks, a certain amount of
sensory information, a limited short-term buffer, and the like; and this
consensus has obtained on both sides of the Atlantic. Both of these ten-
dencies, I should stress, are direct bequests from the Broadbent model of
the human being as a vessel that, like a computer, takes in information and
handles a certain amount of it in a certain amount of time.
Still this general approach to cognition has come increasingly under
critical scrutiny. Some commentators, like Allan Allport (1980) of Ox-
ford, suggest that there are essentially no limitations to the amount of
information that can come in through various sensory channels: on All-
port's account, the primary model of information input should be parallel
(multiple entries at multiple points) rather than serial (one bit after an-
other entering at a single point). Scholars like Richard Shiffrin of Indiana
University have suggested that certain processes can become automatic
and will thereafter exert no significant drain on the organism's ability to
take in or filter novel information. On this view, speed and accuracy
eventually become independent of the number of elements to be proc-
essed (Shiffrin and Schneider 1977). More radical critics, like Benny Sha-
non {1985) of the Hebrew University in Jerusalem, contend that contex-
tual information affects any processing of sensory information: they
question whether any bottom-up or any "outward-inward" model,
which treats "bits of information" apart from their meaning and context
of presentation, can do justice to human cognition. They also challenge
the serial, "one operation at a time" logic of the model. While the stan-
dard linear information-processing mode of thinking has not yet been
abandoned by most researchers, these various chinks in the traditional
picture have led many researchers to rely on parallel, top-down, or con-
textual approaches to the processing of information.
In addition, a closer look at the "modal model" itself pointed up
genuine problems, and the rewards of this research became more elusive.
The more closely the hypothetical stages were examined, the more they
blended into one another; short-term memory could not be readily sepa-
rated from intermediate memory; pre-attentive processes merged with
sensory buffers; processing often did not appear to be strictly serial; the
assumption that specific factors affect specific stages of the sequence
proved difficult to sustain; an individual's expectations might even affect
the early stages of recognition (Glass, Holyoak, and Santa 1979).
Not even the most highly touted experiments withstood close scru-
tiny. While the Sternberg model proved an elegant description of a particu-
123
lar experiment, even slight changes in the model affected human perform-
ance in significant ways. For example, if the probability of test items in the
memory set is unequal, reaction times are faster to more probable items;
if an item is repeated in the memory set, reaction times to it are usually
short; if the items in the memory set are drawn from different conceptual
categories, subjects search exhaustively within categories, instead of
searching every category in the memory set. It has even turned out that
some findings can be accounted for equally well using a parallel (as op-
posed to a serial) processing model (Lachman, Lachman, and Butterfield
1979). Because of these disturbing findings, often linked to the content of
items, few strong generalizations of power could be sustained (Glass,
Holyoak, and Santa 1979; Neisser 1976).
But there was an even deeper problem. Even when results held up
reasonably well, there was increasing skepticism about their actual value.
In daily life, one never encounters such meaningless stimuli under such
controlled conditions; meanings, expectations, contextual effects are al-
ways present and are, more often than not, the dominant consideration.
Information-processing psychologists in the Donders-Stemberg-Sperling
tradition developed increasingly elegant models about effects which did
not prove robust when they were changed in various ways; nor did these
models clearly add up to a larger, more comprehensive picture of how
information is processed under real-life situations. Eventually many of
these researchers themselves abandoned this tradition and went on to
other lines of study.
Reactions to the Standard Information-Processing Paradigms:

The Top-Down Perspective
One line of work has investigated the effects of general (so-called
"real-world") knowledge on approaches to apparently simple linguistic
tasks. In an experimental paradigm pioneered by John Bransford and his
colleagues at Vanderbilt University, a subject hears several sentences. One
is expected to listen carefully because one will be questioned about them.
In an unannounced subsequent part of the study, one is exposed to a new
set of sentences and asked to indicate whether one has heard each particu-
lar sentence before and how confident one is in the correctness of one's
judgment. Bransford and his colleagues found that their subjects' re-
sponses were little affected by the precise wording of the sentences they
initially heard. They seemed to listen, right through the surface wording
of a sentence, to its meaning; were likely to accept a sentence that made
the same point as one of the initial ones; and even combined the contents
of disparate sentences if they fitted comfortably together. Suppose, for
124
instance, that a subject heard the separate sentences: "The ants ate the
jelly," and, "The ants were in the kitchen." Subjects so stimulated tended
to say that they actually heard the sentence, "The ants in the kitchen ate
the jelly," or even, "The ants in the kitchen ate the jelly which was on the
table" (Bransford and Franks 1971, p. 339).
The Bransford paradigm highlights the inferential and integrative
approaches subjects often exhibit in dealing with language, and thus casts
doubt on the legitimacy of conclusions drawn from the hundreds of ex-
periments in which individuals are asked to remember nonsense syllables
or phrases. Rather than suggesting a passive, rote kind of recall, the Brans-
ford results indicate that subjects are actively and constructively process-
ing information and are inferring meaning rather than recalling sheer
strings of words.
A cluster of further studies from the Bransford group fills out this
perspective on "organizing schemas" (Bransford and Johnson 1972; Brans-
ford and McCarrell1975; Bransford et al. 1979; Johnson, Bransford, and
Solomon 1973). Researchers have found that subjects' abilities to process
a paragraph differ dramatically, depending upon whether they have been
provided with a title or a relevant picture to look at beforehand; the title
or picture creates a set that strongly influences how a sentence is inter-
preted. Other studies reveal that subjects inevitably draw inferences about
the sentences they hear, and answer questions based on those inferences
rather than on the literal contents of the sentences themselves; and that
different cues will help an individual remember a given word, depending
upon which sense of that word has been elicited by the initial sentence in
which it was presented. Such supplementing, interpretive, or inferential
activities tum out to be common if not automatic in the processing of
verbal materials; and, in fact, when meaningful nonverbal materials are
used, be they pictures or songs, the same kinds of "organizing schema" can
be discerned as well.
Once we move beyond the level of sentences and approach para-
graphs, stories, or whole texts, the role of organizing schemas becomes ever
more evident. In an influential line of research, David Rumelhart (1975)
and his colleagues have investigated how individuals remember stories.
Building on some of the ideas in Bartlett's earlier work, and also on
Chomsky's analysis of sentences, the Rumelhart circle has put forth the
notion of a story grammar-an underlying set of assumptions about how the
plot of an ordinary story will unfold. A story-grammatical approach posits
that a story has an actor who seeks various goals, takes steps to achieve
these goals, experiences various reactions in approaching these goals, and
eventually fulfills (or fails to fulfill) them. People bring such expectations
to stories: the stories that fulfill these expectations will prove easy to
125
remember; but when stories violate these expectations, people are likely
either to forget the stories or to regularize them (a Ia Bartlett) so that they
become consistent with usual story-grammar assumptions. Moreover,
even when individuals deal with other kinds of experience or text, which
are not storylike but feature a regular sequence of specified events, they
also draw upon organizing scripts or frames: for example, the customary
way in which one changes a tire or visits a restaurant colors one's experi-
ence of, and memory for, new instances of such an incident.
Paralleling trends in other cognitive sciences, this interest in schemas,
scripts, and other inferential and organizing processes has had a major
effect on theory in cognitive psychology. It is not precisely that the belief
in bottom-up or in "sequential serial processing" has been abandoned:
such models still have their uses and adherents. But researchers have come
to appreciate anew that human subjects do not come to tasks as empty
slates: they have expectations and well-structured schemata within which
they approach diverse materials, even including the apparently colorless
and meaningless stimuli of a standard information-processing paradigm.
Thus, an influential alternative approach in cognitive psychology focuses
instead on how the organism, with its structures already prepared for
stimulation, itself manipulates and otherwise reorders the information it
freshly encounters-perhaps distorting the information as it is being as-
similated, perhaps recoding it into more familiar or convenient form once
it has been initially apprehended.
This top-down schematic-based approach stems from earlier efforts in
cognitive psychology: from the "determining tendency" of the Wiirzburg-
ers; from the search for organized structure by Gestalt psychologists; and,
most directly, from Bartlett's story schemas. It also exhibits revealing ties
to Bruner's work on strategies of concept formation. As it happens, the
particular line of work on concept formation undertaken by Bruner, Good-
now, and Austin has rarely been followed up-probably because the view
of concepts as arbitrary clusters of features has been as thoroughly rejected
in cognitive psychology as have Ebbinghaus's nonsense syllables (more on
that in chapter 12). However, interest in the strategies or the organizing
schemes that one brings to a task is a robust legacy of the Bruner enter-
prise.
The adherents of this approach to cognition feel that it is on firm
ecological grounds. Where the users of artificial stimuli and arbitrary tasks
need to prove the relevance of their approaches to the real world, those
who adopt the contextual, top-down approach can point to the fact that,
in everyday experience, such highly interpreted and meaningful stimuli are
the norm.
Even within the relatively conservative area of memory research, there
126
has been a shift in the perspective of many researchers. Instead of the
Ebbinghaus intoxication with meaningless information, researchers use
meaningful and even highly elaborated, context-rich materials, like stories
(Bower 1976). Attention is now being directed to the "depth" with which
the information is processed. On this view, a subject has the option of
paying attention only to superficial aspects of the stimulus (say, the sounds
of words or the precise syntactic form of phrases) or of assimilating it to
various schemata that have already existed: the more information is en-
veloped in earlier ways of knowing and embedded with rich associations,
the deeper the level of processing; and hence, the more likely that the
information will be firmly encoded and adequately remembered. Whether
the stimulus is processed at a shallow (surface or sensory) level, or at a
more semantically integrated level, depends on the nature of the stimulus,
the time available for processing, and the subject's own motivation, goals,
and knowledge base.
Such a "depth of processing" approach (Craik and Lockhart 1972) has
at least partially replaced sequential models of information processing. On
this view, the nature of incoming information is determined by the opera-
tions performed during its input. Input operations include analysis not
only of sensory aspects but of semantic features as well. Thus, memory
depends on the nature of the code and the type of analysis undertaken, and
not on the properties of particular memory stores. Moreover, one form of
encoding is not necessarily superior to another. Ordinarily it is preferable
to perform deep processing. However, it turns out that a more superficial
mode of processing is recommended if, for example, one is trying to re-
member which words rhyme with an initial list (Lachman, Lachman, and
Butterfield 1979, p. 279). Ultimately, psychology's contemporary memo-
rizer has many options: one can choose to process information at different
depths; and one may even find that, for some purposes, a more shallow
mode of processing is preferable. We have come a long way from the
compulsory stream leading from periphery to long-term memory. Indeed,
in this concern with goal-directedness, diverse approaches to information,
and highly meaningful contents, psychology is re-embracing parts of the
Wiirzburg program.
An interest in peripheral forms of stimulation and in context-free
processing is certainly justifiable, but all too often it seems to leave the
heartland of human psychological processing unscathed. Proceeding in a
molar and top-down fashion, this work on schemas, stories, and depth of
processing strikes many observers (including me) as closer to what psy-
chology ought to focus upon. And yet the question arises whether this
strand of psychology has achieved much that was not evident to our
predecessors or, indeed, to nonpsychologically trained observers. After all,
127
it hardly takes twentieth-century technology and high-power statistics to

demonstrate that individuals bring certain experiences and organizing
frameworks to a new activity, or that it is possible to process information
with more or less effort for sundry purposes. And if we embrace a twen-
tieth-century approach, the question arises whether we have advanced
beyond the Gestaltists and the Wiirzburgers or are simply rediscovering
what they knew (and what the narrow behaviorists and information
processors had apparently forgotten).
There is no easy answer to this question. In my view, the major
advance over previous work inheres in the better understanding of which
structures individuals bring to experience, and of which manipulations are
likely to cause a significant change in processing, memory, or inference; in
other words, it is in the fine details of how such top-down process works,
and not in the fact of its existence, that psychological advances consist.
While less exciting than the discovery of an entirely new set of processes,
and scarcely counterintuitive, this understanding is, to my way of think-
ing, no less significant. At times, advances of science occur in the details
of understanding, rather than in the promulgation of revolutionary new
laws. This may be the fate of psychology in our time.
Mental Representations
As the molar approach continues to struggle with the molecular ap-
proach, other lines of investigation bear in critical ways on the classical
information-processing model of the 1950s and 1960s. That model was
essentially content-blind: the assumption was that information of any sort is
processed in essentially the same manner. Thus, if short-term memory
holds seven slots, these slots are equally capacious for any kind of informa-
tion, be it verbal, pictorial, or musical, be it dross or gold. By the same
token, if a search takes a certain period of time per item and is exhaustive
rather than self-terminating, these processes obtain irrespective of what is
being searched for and of who is doing the searching.
These assumptions, which were made by all the pioneering cogniti-
vists on both sides of the Atlantic, served as simplifying beginning points.
They also reflected the computer model and information-theoretical ori-
gins of modem cognitive psychology: with such mechanisms, it is rela-
tively easy to set up experiments so that actual contents are irrelevant.
After all, information theory and computers are deliberately constituted to
be content-blind.
Recently, however, these assumptions have been called into question.
One group of scholars, represented by Roger Shepard of Stanford Univer-
128
sity, has been exploring the operation in individuals of a form of represen-
tation involving visual imagery. Shepard has studied the abilities of in-
dividuals to form mental representations of objects-be they familiar or
unfamiliar shapes-and to answer questions about their physical similari-
ties. In typical studies, subjects are asked to judge whether two presented
geometrical forms are the same when one of them has been rotated, or to
answer questions about other kinds of entities, which can be imaged, such
as the shapes of the fifty United States (Shepard and Chipman 1970).
With his colleague Jacqueline Metzler, Shepard discovered that reac-
tion times preceding decisions about the identity of forms directly reflect
the size of the angle of difference between the orientation of the two forms
(1971). In cases where the angle between the figures is close to zero, a
response is given almost instantaneously: as the angle climbs toward 180
degrees, a subject's reaction times climb in a linear relation to the size of
the angle. The authors interpret such findings as evidence that the subjects
are actually effecting the comparison by mentally rotating one or another
figure at a constant rate. Such mental imagery mimics the trajectory
through which the figures would pass if they were actually in hand and
being rotated before one's eyes.
These experiments, as carried out by Roger Shepard and (in recent
years) by Stephen Kosslyn, are controversial but I shall postpone discus-
sion of such debate to chapter 11. For now, what is germane is Shepard's
belief that cognitive scientists have erred in positing propositional (lan-
guagelike) representations as the lingua franca of cognitive systems. The fact
that computers can-and usually do-transmit information in only one
symbolic form is no reason to assume that human beings do the same.
Shepard sees mental imagery as a human capacity that has evolved over
millions of years to enable the organism to deal with an ever-changing
environment whose consequences it must be able to anticipate. Whatever
the initial focus, such knowledge has been internalized through evolution
so that it is now "pre-wired" in individuals and governs how they appre-
hend objects in space.
Nor are these capacities limited to the visual modality. Shepard's work
with audition suggests that the representation and transformation of musi-
cal sounds reflect the same general principles as does visually presented
information: principles of conservation, symmetry, proximity, and the
like. Thus, Shepard comes to endorse some of the underlying notions of
Gestalt psychology while placing them on more rigorous experimental
footing (Shepard 1981, 1982; see also Jackendoff 1983).
Though growing out of research on information in a single sensory
modality, Shepard's work challenges much of cognitive-psychology main-
stream. If he can make a convincing case for more than one mode, the
129
question arises whether there might not be a multiplicity of modes of
mental representation, each tied to a particular content. In the interests of
parsimony, most philosophers and not a few psychologists have voted in
favor of a single form. The positing of another form of mental representa-
tion complicates the picture-for if two representational modes, why not
three, or seven, or three thousand? From Shepard's perspective, however,
it is ill-advised to adjust the findings of human experimentation to the
demands of the digital computer.
Psychology's Contributions
In my view, cognitive psychology has made impressive progress since the

mid-1950s, inventing paradigms that are widely and productively used,
and investigating a variety of topics, of which I have named but a few.
Cognitive psychologists have identified fascinating phenomena, ranging
from the number of units that can be held in mind at any one moment to
the manner in which geometric forms "are mentally manipulated" by
normal adults; they have initiated many intriguing comparisons, ranging
from the difference between concrete and formal operations in children, to
the contrast between propositional representation and visual forms of
imagery; they have devised a slew of new methods, some technically
sophisticated, while also refining techniques that have been around for a
century. Certainly, as William Kessen has put it (1981), several interesting
new settlements have been founded and are being populated; and re-
searchers have become expert in such areas as mental imagery, conserva-
tion of liquid, or the use of schemas for understanding stories.
Perhaps even more important, cognitive psychologists have won the
battle on their chosen field within psychology. While behaviorism still has
appeal as a method for dealing with various clinical populations, its theo-
retical superstructure and its experimental strictures no longer exert much
influence within the research community. Nearly all researchers accept the
need-and the advisability--of positing a level of mental representation.
These scholars deal comfortably with concepts like schemas, mental opera-
tions, transformations, and images. Disputes center on the best model of
representation, not on the need for some such conceptual apparatus.
But while cognitive psychology emerges on these criteria as a singular
success, substantial problems remain-some within the field, others hav-
ing to do with the place of this discipline within a larger cognitive enter-
prise. A first evident problem is the splintering of the field. There are now
many specialties and subspecialties, each proceeding more or less well on
its own, but having relatively little contact with neighboring work. Re-
130
search on visual perception makes little contact with research on story
understanding or musical memory, and often the concepts used to account
for these phenomena are equally remote from one another.
A forthright and ambitious effort to introduce unifying constructs
into cognitive psychology comes from the work over the past decade of
John Anderson (1983). This psychologist, deeply rooted in the practice of
artificial intelligence, has developed the so-called ACT (for "Adaptive
Control of Thought") system which is put forth as a general model of the
"architecture of cognition."
ACT incorporates a process model that describes the flow of informa-
tion within the cognitive system. The central notion is a production sys-
tem (see chapter 6 for further discussion). As soon as a node in a network
is sufficiently activated, a certain action (or production) is carried out: this
construct has been described as a kind of cognitive stimulus-response
bond, because whenever the proper stimulating phenomena are present, an
action is elicited. The system itself includes different kinds of memory-
a working memory (consisting of information with which the system can
currently work), a declarative memory (with propositions in it), and a prOduction
memory (which involves the actions carried out by the system). There are
numerous other mechanisms as well. Encoding processes deposit informa-
tion about the outside world into working memory. Performance processes
convert commands in working memory into behavior. A storage process
creates permanent records in declarative memory; a retrieval process re-
trieves information from declarative memory; and an execution process
deposits the action of matched productions into working memory. A
match process places data in working memory in correspondence with the
conditions of production. Anderson's theoretical work involves specifying
the nature of the knowledge structures that reside in working memory, the
storage process, the retrieval process, and the various factors that activate
productions.
As should be evident from this barrage of mechanisms and concepts,
Anderson's system is complex, as is the evidence he has brought to bear.
The system is also controversial, with some psychologists putting their
faith in the kind of enterprise in which Anderson is engaged, and others
suspecting that it is built on a base of sand. As Keith Holyoak, not an
unsympathetic critic, has noted:
Some who have offered empirical evidence apparently contradicting [Ander-

son's] theory have been frustrated in their attempts to hit a moving target .... Some
have worried that ACT is less a predictive theory than a very general framework
that imposes few detailed constraints on the nature of the cognitive system. And
the most skeptical critics have claimed that the search for a general theory of
cognition is simply a wasteful misdirection of energy. (1983, p. 500)
131
I I I THE CoGNITIVE SciENCES : HISTORICAL PERSPECTivE
My own feeling is that it is premature to call off efforts to locate and

describe a general cognitive system. Like Piaget and other "systematizers,"
Anderson is bold enough to attempt such a daunting task; and, even if he
fails, we should learn from the kinds of moves (and the mistakes) he
makes. I find, however, one telling difference between Piaget and Ander-
son. Whatever Piaget's theoretical meanderings, he always had a deep
involvement with the phenomena of cognition and drew his principles
from intensive work with children in a rich clinical setting: these are, in
fact, his most lasting contributions to psychology. At times, Anderson
seems to have a second-order psychology: a psychology based on reading (and
sometimes taking too seriously) the psychological treatises of other schol-
ars, and on attempts to do psychology in a way that fits into the digital
computer, rather than on wide immersion in the actual phenomena of
thinking, remembering, perceiving, and the like. To the extent that this
characterization is apt, it suggests that ACT may be internally coherent
rather than externally linked to the actual flowing processes of human
thought.
There has been a long tradition within psychology which ignores the
particular content of information-a tradition that dates back to the work
of Donders and that was fully subscribed to in the pioneering researches
of Broadbent, Bruner, Miller, and their peers. Anderson's work fits com-
fortably into this tradition. However, in recent years, several workers, and
most especially Jerry Fodor (1983), have suggested that the mind is better
construed as a number of largely separate information-processing devices,
including ones constructed to deal with language, visual processing, music,
and other such specific kinds of content. On this "modular'' view, modules
have evolved to carry out specific forms of analysis in a hard-wired, rapid,
and encapsulated manner; communication between the modules occurs
only subsequently, in ways that remain obscure. There is also considerable
skepticism among modularists on the need for some kind of central proces-
sor or executive function; and even those who (like Fodor) believe in such
central systems are skeptical that cognitive science will ever be able to
account for them.
The modular point of view, for which I have considerable sympathy
(Gardner 1983), stands in contrast to those cognitive theories that are
"content blind" as well as to those that subscribe to a belief in "horizontal
faculties." On a horizontal view of the sort embraced by Anderson, facul-
ties like learning, memory, and perception are assumed to work in the same
or similar fashion, independent of whether the content is verbal, pictorial,
musical, gustatory, or the like. But as Fodor and I have attempted to show,
there is accumulating evidence that "vertical" psychological mechanisms
deal in individual fashion with different contents. It appears probable that
132
the ways we humans deal with, say, syntax in language share few funda-
mental properties with our transformations of spatial images or our inter-
pretations of musical expressiveness. None of these possibilities were ade-
quately appreciated by the first generation of psychologist-cognitivists,
and they have been basically ignored or minimized by those who, like
Anderson or Piaget, attempt to produce general cognitive architectures.
If the modularists are on the right track, there is a disturbing possibil-
ity for psychology. Rather than being a single coherent discipline-as its
leading figures have understandably wished-psychology may tum out to
be more of a holding company. In such a vision, there will be separate
studies of language, music, visual processing, and the like, without any
pretense that they fit into one supradiscipline. (In my concluding chapter,
I suggest that this may be a possible fate for cognitive science as a whole.)
Since this outcome would fly in the face of the established wisdom within
the field, it is important to continue to put forth synthetic accounts as
scholars like Anderson are doing.
While current debates pit the generalists against the modularists (see,
for example, Piattelli-Palmarini 1980), it may be that both the modularists
and the central processors have hit upon important truths. The modularists
may be right in thinking that many domains operate by their own laws;
the centralists, in believing in another, synthetic intellectual realm where
modular processes are inadequate, and where horizontal processes prove
necessary. Debate might then come to center on whether any of the modu-
lar realms can be subsumed under some aspect of a central-processing
view.
Discussions of the modular as against the centralist approach are
conducted by those who are generally in sympathy with the current meth-
ods and concepts of cognitive psychology. More severe criticisms of the
enterprise have also been leveled, however. One school deplores the artifi-
ciality of studies in cognitive psychology. In this view, the most convincing
models pertain to tasks that have little clear-cut relation to ongoing human
activity; in contrast, those studies that do pertain to ongoing human activ-
ity take the form of intriguing demonstrations rather than theory-enrich-
ing experiments. In its harshest form, this criticism harbors the suggestion
that the whole information-processing approach has been misguided; that
attempts to model human cognition, in terms of a series of operations upon
mental content, is simply a bad model of the human mind; and that some
radically different (though perhaps not yet articulated) approach will be
needed (Shanon 1984). I shall return to this theme when I consider the
computational paradox.
Such skepticism has even been voiced by some of the leading experts
in psychology. In his 1967 book Cognitive Psychology, Ulric Neisser exuded
133
I I I THE CoGNITivE SciENCES : HISTORICAL PERSPECTIVE
optimism concerning the promise of the field in which he had been work-
ing. But a decade later in Cognition and Reality (1976), he was much more
sober. Berating his colleagues for both their homage to the computer model
and their insistence on artificial laboratory scenarios, he lamented the lack
of an ecologically valid psychology-one that speaks to the kinds of issue
human beings encounter and resolve in everyday life.
As Neisser described it, such a psychology-lacking ecological valid-
ity, indifferent to culture-risks becoming a narrow and uninteresting
specialized field. And, in his view, "the villains of the piece are the me-
chanistic information-processing models, which treat the mind as a fixed-
capacity device for converting discrete and meaningless inputs into con-
scious percepts" (1976, pp. 7, 10). Neisser (1984) calls for a perceptual
psychology, which studies how humans see while moving around and
interacting with objects (Gibson 1979); for concept-formation research,
which features complex objects in the world (Rosch 1973b ); and for a study
of memory which includes autobiographical accounts, eyewitness reports,
and memory of childhood friends (Bahrick, Bahrick, and Wittlinger 1975).
John Morton, one of the leading information-processing psychologists in
Britain, has declared, "Experimental psychology has a disastrous history
with respect to its relevance" (1981, p. 232). And Arnold Glass, Keith
Holyoak, and John Santa, authors of a well-regarded textbook in cognitive
psychology, raise the unsettling possibility that psychology is a "fast race
on a short round track." They explain that this remark is meant to imply:
Cognitive psychology is not getting anywhere; that in spite of our sophis-

ticated methodology, we have not succeeded in making a substantial contribution
toward the understanding of the human mind.... A short time ago, the informa-
tion processing approach to cognition was just beginning. Hopes were high that the
analysis of information processing into a series of discrete stages would offer
profound insights into human cognition. But in only a few short years the vigor
of this approach was spent. It was only natural that hopes that had been so high
should sink low. (1979, p. ir)
While Glass and his colleagues do not themselves endorse this bleak out-
look, they correctly note that this view is expressed "with depressing
frequency these days," and, as I have suggested, outside as well as within
the halls of cognitive psychology.
In my own view, there was understandable but probably excessive
excitement about cognitive psychology during the first few years following
its birth (or rebirth). Not surprisingly, then, some of those who were
jubilant at the demise of behaviorism have been less than ecstatic at the
results of the last (or first!) twenty-five years. There is certainly nothing
wrong with experiments conducted under artificial conditions per se (see
134
the example of physics!), but there is something bothersome when findings
prove fragile once the experimental conditions are altered slightly, applied
to more complex phenomena, or moved one step outside the laboratory.
Part of the dissatisfaction at the present time comes, paradoxically,
from the perfection of experimental methods within psychology. One of
the genuine contributions of behaviorism in psychology was the refine-
ment of experimental methodology; but in this case, the refinements may
have reached the point of diminishing returns. As I read the literature, too
much emphasis has been placed on having an experimental procedure
without any perceptible flaws or ambiguities; and all too often this empha-
sis takes place at the expense of considering what is an interesting or
important problem. By the same token, once an interesting demonstration
has been introduced into the psychological literature, dozens of other
experimenters devote their attention-and their methodological zeal-to-
ward finding the vulnerabilities in the experiment. And ultimately, nearly
all these experiments get shown up (and ultimately abandoned) for their
limitations. These trends combine to produce a science that has progressed
rather less than it should have, and that still consists more of a set of
impressive but isolated findings than of a truly cumulative discipline.
It hardly needs saying that it has been easier to achieve methodolog-
ical perfection in those studies that deal with molecular phenomena and
proceed in a bottom-up fashion. As I have noted, however, there recently
has been growing dissatisfaction with experiments in this tradition: and
there has been a correlative attraction for psychological investigations that
deal with more molar phenomena (like the understanding of a story or the
transformation of a mental map) and that do so in a top-down fashion
(taking into account schemas, frames, and the surrounding context). The
pendulum has definitely been swinging in a molar direction once again.
These two situations-the intoxication with method and the trend
toward molarity-provide psychology with an invaluable opportunity.
Put simply, it is time for psychology to wed its indubitable methodological
sophistication to a concern for problems that are more molar, less artificial,
more representative of real-life situations, more substantive. Just how to
accomplish this is hardly an easy question to answer-though I have tried
to indicate some of the lines of investigation that I personally find most
promising. In this context, I strongly believe that psychologists can benefit
from studying the example of artificial intelligence. In spite of (or perhaps
because of) the presumed rigor of the computer, researchers in the A.l.
tradition have been much more ready to deal with molar topics (like
problem solving or story understanding) and to do so in a top-down
fashion. The partnership that is developing between the two fields-and
that will occupy me more fully in the next chapter-may serve as the best
135
II I THE CoGNITivE SCIENCES : HISTORICAL PERSPECTIVE
model for how to wed methodological sophistication to problems of

unquestioned substance.
Discussion of these interdisciplinary matters brings me to the remain-
ing issue of whether, like philosophy, psychology will remain a viable
discipline, or whether it will eventually be absorbed into another discipline
or set of disciplines. To my mind, psychological methods will always be
central in any cognitive science enterprise. Indeed, it is in the perfection
of methods for studying individual cognition, for comparing groups, or for
tracing processes over time that psychology has probably realized its most
outstanding achievements. It is simply inconceivable to me that studies of
music, language, or visual spatial abilities will be carried out without the
techniques and approaches honed over the past century by psychological
researchers.
Psychology is central to any cognitive science and thus is especially
likely to be absorbed by a larger field-in contrast to, say, philosophy or
neuroscience. Philosophy may contribute an agenda of questions to cogni-
tive science, but it contributes agendas to many other sciences as well and
thus should continue to exist whatever good or ill befalls cognitive science.
On the other hand, neuroscience with its well-defined domain-the study
of the human brain and the nervous system-can also continue irrespective
of events in psychology or cognitive science.
To the extent that the mind is indeed a single coherent domain-as
Wundt, Piaget, and John Anderson believed-it will make sense for psy-
chologists to continue to practice their calling under the same label and
with the same goals in mind. If, however (as I believe is the case), the
terrain of cognition is better mapped in terms of certain problems or topics
-such as visual perception, object classification, rationality, conscious-
ness, and the like-then cognitive psychology may lack a central subject
matter, a "natural domain."
I believe that it is only as part of a research team that cognitive
psychology is likely to survive. Psychological methods (observational as
well as experimental) need to be brought to bear in the analysis of a field
like language (in cooperation with a linguist), or music (in cooperation
with a musicologist), or problem solving (in cooperation with a logician,
a technician, or a physical scientist). The psychologist can then test the
distinctions offered by the domain expert; and programs of artificial intelli-
gence can be devised to indicate whether the models proposed by the
psychologist-domain-expert team are actually viable. I anticipate the
merger of cognitive psychology and artificial intelligence into the central
region of a new unified cognitive science, with fields like linguistics, music
theory, or spatial analysis providing the appropriate frameworks for prob-
lems that lie within their domain.
136
While I expect that cognitive psychology will simply, and advanta-
geously, be absorbed into an issue-oriented cognitive science, I do not
mean to indicate that psychology as a whole will disappear. (Far be it for
me to predict-or wish-the end of my own discipline!) There will always
be many practical needs for psychologists in the clinic, in the school, and
in industry. Even on the level of theory, some pockets of psychology will
endure. I do not anticipate that topics like human personality or motiva-
tion will be absorbed by other disciplines: neither neuroscience nor artifi-
cial intelligence can gobble them up. By the same token, psychology will
continue to be indispensable in illuminating the differences among human
beings: between normal and exceptional individuals, between individuals
with different kinds of disease, or individuals from different social or
cultural settings. The practice of individual or differential psychology
should continue and may well yield interesting scientific principles. By the
same token, psychology can play an equally valuable role in effecting other
kinds of comparison-between one sensory modality and another, be-
tween one symbol system and another, between one form of representa-
tion and the other-whether or not, in the end, the modularity view is
judged as more powerful than the central-processing view.
Psychology has been one of the great disciplinary success stories of
this century, and it seems inopportune to be predicting its disappearance
into a larger discipline. In fact, however, it is precisely the innovations in
psychology of the past few decades which have given rise to a larger
cognitive revolution, within which psychology has clearly been a central
discipline. In no ways should psychology be considered simply a holding
operation until neurology or sociology or anthropology comes along: as
much as any scholars, psychologists have successfully made the case for
the centrality of the human mind and mental representation within the
current scientific milieu. The ultimate merger of psychology with artificial
intelligence and other disciplines into a larger cognitive science will be a
tribute to its success and its significance.
Perhaps events of the past century have demonstrated that psychol-
ogy is a difficult science to bring to completion, at least without the aid of
other disciplines. But even more clearly, we have come a long way since
Kant declared, less than two centuries ago, that a psychological science was
impossible.
137
6
Artificial Intelligence:
The Expert Tool
I am prepared to go so far as to say that within a few years,
if there remain any philosophers who are not familiar with
some of the main developments in artificial intelligence, it
will be fair to accuse them of professional incompetence,
and that to teach courses in philosophy of mind, episte-
mology . . . without discussing . . . aspects of artificial
intelligence will be as irresponsible as giving a course in
physics which includes no quantum theory.
-AARON SLOMAN
The Summer of 1956 at Dartmouth
In the summer of 1956, a group of ten young scholars trained in mathemat-

ics and logic gathered on the campus of Dartmouth College in Hanover,
New Hampshire. Their purpose: to confer about the possibilities of pro-
ducing computer programs that could "behave" or "think" intelligently.
As they had declared in their grant application to the Rockefeller Founda-
tion: "The study is to proceed on the basis of the conjecture that every
aspect of learning or any other feature of intelligence can in principle
be so precisely described that a machine can be made to simulate it'"'
(McCorduck 1979, p. 93).
Of the numerous scholars who attended parts of the summer institute,
four in particular came to play crucial roles in the development of a new
field called artificial intelligence. First of all, there was John McCarthy,
then an assistant professor of mathematics at Dartmouth and eventually
*This historical sketch follows the account given by Pamela McCorduck (1979), a
historian of artificial intelligence.
138
Artificial Intelligence: The Expert Tool
the founder and first director of the A.I. labs at both the Massachusetts
Institute of Technology (1957) and Stanford University (1963). McCarthy
was the major organizer of the institute and the coiner (according to most
accounts) of the term artificial intelligence. The remaining three leading figures
were Marvin Minsky, then a Junior Fellow in mathematics and neurology
at Harvard and eventually the director of the Artificial Intelligence Labora-
tory at M.I.T.; and Herbert Simon and Allen Newell, then at the Rand
Corporation in Santa Monica and also at Carnegie Institute of Technology
(now Carnegie-Mellon University) in Pittsburgh, where they have re-
mained until this day.
The summer at Dartmouth gave these and other scholars a chance to
exchange views and to arrange to collaborate on future work. Various
authorities during the 1940s and early 1950s had expressed the belief that
computers should be able to carry out processes resembling human think-
ing, and it was the job of the present assemblage to put these promises to
the test. Alex Bernstein, then a programmer for International Business
Machines in New York City, talked about the chess-playing program on
which he was working. Arthur Samuel, also of the I.B.M. Corporation,
discussed a program that played checkers. Newell and Simon described a
program that they had devised to solve theorems in logic. Nathan Roches-
ter of I.B.M. in Poughkeepsie described work on programming a model of
neural networks, while Marvin Minsky discussed the use of computers to
prove Euclidean theorems.
The meeting at Dartmouth did not fulfill everyone's expectations:
there was more competition and less free exchange among the scholars
than the planners had wished. Nonetheless, the summer institute is con-
sidered pivotal in the history of the cognitive sciences, in general, and in
the field of artificial intelligence, in particular. The reason is, I think, chiefly
symbolic. The previous decade had seen the brilliant ideas of an older
generation-Norbert Wiener, John von Neumann, Warren McCulloch,
Alan Turing-all point toward the development of electronic computers
which could carry out functions normally associated with the human
brain. This senior group had anticipated developments but were uncertain
whether they themselves would have the opportunity to explore the
promised land.
At Dartmouth, members of a younger generation, who had grown
up in an atmosphere seeded with these seminal ideas, were now ready
(and in some cases, beyond mere readiness) to devise the machines and
write the programs that could do what von Neumann and Wiener had
speculated about. These younger scholars were attracted by powerful (if
still vague and poorly understood) notions: data being processed by a
program and then becoming part of the program in itself; the use of
139
computers to process symbols rather than simply to "crunch numbers";

the proclivity of new languages for bringing out unsuspected potentials
in the machine's hardware; the role of computers in testing scientific
theories. Perhaps conducted in isolation, the meeting might not have had
much of an impact. But it came just weeks before the meeting at M.I.T.
(see pages 28-29), where some of the same participants, and such formi-
dable figures from neighboring fields as Noam Chomsky in linguistics
and George Miller in psychology, were presenting their own ideas to the
emerging world of cognitive science. And, finally, it was the time as well
of key publications-not only Simon and Newell's Logic Theorist and
Marvin Minsky's widely circulated "Steps toward Artificial Intelligence"
(1963), but also important monographs by Bruner, Chomsky, Levi-
Strauss, Piaget, and many other scholars of a cognitive bent. While no
single event can lay claim to signaling the "birth" of all of cognitive
science, the workshop at Dartmouth is the chief contender within the
field of artificial intelligence.
The Ideas of Artificial Intelligence
Since the Dartmouth meeting, artificial intelligence has had a brief but
stormy history. Part of the storm swirls around definitions. Nearly all
authorities agree that artificial intelligence seeks to produce, on a com-
puter, a pattern of output that would be considered intelligent if dis-
played by human beings. Most authorities see the computer program as
a test of a particular theory of how a cognitive process might work. But
thereafter consensus falters. Some definitions stress the devising of pro-
grams; others focus on programming languages; others encompass the
mechanical hardware and the human conceptual component, as well as
the software. Some practitioners want to simulate human thought pro-
cesses exactly, while others are content with any program that leads to
intelligent consequences.
Authorities also disagree on how literally to take the thinking meta-
phor. Some researchers take what has come to be termed the "weak view,"
where the devising of "intelligent" programs is simply a means of testing
theories about how human beings might carry out cognitive operations.
Others, however, put forth much more forceful claims about their field.
According to the view of "strong AI," as phrased by philosopher John
Searle, "the appropriately programmed computer really is a mind, in the
sense that computers given the right programs can be literally said to
understand and have other cognitive states. In strong AI, because the
programmed computer has cognitive states, the programs are not merely
140
Artificial lnfelligence: The Expert Tool
tools that enable us to test psychological explanations; rather the programs
are themselves the explanations" (1980, p. 417). I shall consider the merits
of weak and strong A.I. at the conclusion of this chapter.
But while the tension between weak and strong claims is one of the
most momentous debates, it is by no means the only one. As Robert
Wilensky, a leading artificial intelligence researcher, recently commented,
"Artificial intelligence is a field renowned for its lack of consensus on
fundamental issues" (1983, p. xii). Indeed, in a recent capsule history of
artificial intelligence, Allen Newell (1983) was able to single out no fewer
than three dozen issues that have at times divided the field. While some
of these are quite technical, and others of only transient interest, two of
them seem to me of particular note. The first is the tension between
"generalists" and "experts"-a tension recalling the dialectic between the
modular and the central-processing perspectives in contemporary psy-
chology. Generalists believe in overarching programs (or families of pro-
grams) that can be applied to most any manner of problem: experts place
their faith in programs that contain much detailed knowledge about a
specific domain but prove relatively restricted in their applicability. A
second tension has to do with the scientific status of the field. While
some of the founders were prepared to make strong claims for scientific
importance (and, indeed, see A.I. as replacing epistemological pursuits),
more skeptical commentators have wondered whether artificial intelli-
gence deserves to be considered a scientific discipline at all. From their
point of view, A.I. is simply a form of applied engineering-even gim-
mickry-with no real standing as a theoretically based scientific disci-
pline. To be sure, similar skeptical challenges have been leveled at other
of the cognitive sciences; but perhaps because of the dramatic promise of
a "thinking machine," the battles about the scientific status of A.I. have
been particularly vehement.
In this chapter, I shall observe the wide swings of mood and the
diverse viewpoints that have characterized the leading practitioners and
commentators during the first three decades of A.I. It is not possible, of
course, to touch on every strand of artificial intelligence; for example,
except incidentally, I shall not discuss work on robots, retrieving informa-
tion from data bases, planning optimal combinations or optimal schedules,
simulating organizational activities, or writing programs that write pro-
grams, even though each of these areas is becoming part of standard
reviews of artificial intelligence (Nilsson 1980; Waldrop 1984a, 1984c;
Winston 1977). But I will touch on those lines of work that seem to me
to be most relevant to human psychology and, at the conclusion of this
chapter, attempt to situate the field of A.I. within the broader framework
141
The Dream of Artificial Intelligence

A.I. may have a short history, but the dream of a mechanical mind
goes back a long time. Early intimations can be discerned in the work of
Rene Descartes, who was interested in automata that could simulate the
human body (he was skeptical about simulating the mind). Whether or
not inspired by Descartes, thinkers within the French tradition seem to
have been the first to pursue the idea of a machine that can reason. In
Paris in 1747, a French physician named Julian Offray de la Mettrie pub-
lished his book L Homme Machine, in which he argued that "the human
body is a machine that winds up its own springs" (quoted in McCorduck
1979, p. 36), and that the brain, as the organ of thought, was subject to
study and duplication. As he put it, "thought is so little incompatible
with organized matter, that it seems to be one of its properties on a par
with electricity, the faculty of motion, and impenetrability" (quoted in
Lowry 1971, p. 42).
Pursuing such a train of thought on a more practical level was the
craftsman Jacques de Vaucanson, a builder of automata who thrilled
Europe during the early part of the eighteenth century with mechanical
flute players, ducks, and tabor-pipe players. In a lengthy accompanying
document, Vaucanson indicated how each part of the flute player was in
fact modeled after comparable components of the human model. Accord-
ing to the historian of psychology John Marshall, Vaucanson "was con-
cerned to formulate and validate-in the most precise and formal language
available to him-a theory of the German flute player" (Fryer and Marshall
1979, p. 261).
The scene now shifts to nineteenth-century England, to investiga-
tors who pursued the mechanization of thought in ways much closer to
our own. One such character was the brilliant and prescient Cambridge
mathematician Charles Babbage who devoted many years to devising an
automatic table calculator which could carry out the complicated compu-
tations needed for navigation and ballistics. Unfortunately the machine
that he designed would have required the production of thousands of
precision parts; and while scientists of today think that Babbage's ma-
chine would have worked, the British government withdrew its support
after investing the then large sum of seventeen thousand pounds. Mean-
while, inspired by his collaboration with one Lady Lovelace, Babbage
became even more grandiose, conceiving of a machine that could tabu-
late any function whatever and could, in principle, play chess (McCor-
duck 1979, pp. 25-27). This "difference machine" (as it was called) was
based on the difference tables of the squares of numbers. It would use
punched cards of the sort hitherto used to control special weaving looms:
142
there were operation cards, which directed the operations to be per-
formed, and variable cards, which determined the particular variables on
which the operations were to be performed (Dorf 1974). Any arithmetic
problem could be set; and, provided that the proper cranks were turned,
the right answer should issue forth.
While Babbage was attempting to implement his ambitious mechani-
cal aspirations, another British mathematician, George Boole of Queens
College Cork, was involved in a different but equally momentous under-
taking: that is, to figure out the basic laws of thought and to found them
on principles of logic. In order to eliminate the ambiguities of natural
language (which had dominated logic since the time Aristotle had studied
the syllogism), Boole used a set of arbitrary symbols (a, b, x, y, and so on)
to stand for components of thought. As he put it, "a successful attempt to
express logical propositions by symbols, the laws of whose combinations
should be founded upon the laws of the mental processes which they
represent, would, so far, be a step toward the philosophical language"
(quoted in Hilton 1963, p. 163). These symbolic elements could be com-
bined or dissociated through operations like adding, subtracting, or multi-
plying, so as to form new expressions, or new conceptions, involving the
same elements. These procedures amounted to a kind of "mental algebra,"
where reasoning could be carried out in abstract positive or negative terms,
unsullied by the particular associations tied to specific contents. And these
operations were termed by Boole the "laws of thought." Most important
for the future, Boole observed that his logic was a two-valued or true-false
system. Any logical expression, no matter how complex, could be ex-
pressed either as 1 (standing for "all," or "true"), or as 0 (standing for
"nothing," or "false"). The idea that all human reason could be reduced
to a series of yes or no discussions was to prove central for the philosophy
and the science of the twentieth century.
The significance of Boole's work was finally appreciated a half-cen-
tury later by Alfred North Whitehead and Bertrand Russell when they
produced their Principia Mafhemafica (1910-13). The goal of this work, as I
have noted, was to demonstrate that the roots of mathematics lie in the
basic laws of logic. Whitehead and Russell relied heavily on the formalism
pioneered by Boole. Russell went so far as to declare in his ascetic manner,
"Pure mathematics was discovered by Boole in a work he called 'The Laws
of Thought'" (quoted in Halacy 1962, p. 106).
The cluster of ideas represented by Babbage's calculating machines,
Boole's laws of thought, and Whitehead and Russell's decisive demonstra-
tions were eventually to be integrated by scholars in the 1930s and 1940s.
Their work culminated in the first computers and, eventually, the first
programs that can be said to display intelligence.
143
Realizing the Dream

Many individuals set the groundwork for the mid-century explosion
which led to Dartmouth and its aftermath. Of tremendous significance was
the work of the M.I.T. mathematician Claude Shannon. In 1938, Shannon
published"A Symbolic Analysis of Relay and Switching Grcuits"-possi-
bly the most important, and also the most famous, master's thesis of the
century. In this work, Shannon showed that relay and switching circuits
of the type found in an electronic machine could be expressed in terms of
equations of a Boolean sort: for the true-false system could parallel "on-off
switches," or closed and open states of a circuit. Indeed, any operations
that could be described in a finite set of steps could be carried out by such
"switching" relays. Shannon's work laid the groundwork for the construc-
tion of machines that carried out truth-logic operations, and also suggested
new ways in which the circuits could be designed and simplified. At a
theoretical level, he also indicated that the programming of a computer
(laying out a set of coded instructions to be precisely followed) ought to
be thought of as a problem of formal logic rather than of arithmetic, an
insight that grew out of Boole's work. In one swoop, Shannon had injected
a subject of purely academic interest into the world of practical machinery,
including the newly emerging computing machines.
Shannon's insights did not occur in an intellectual vacuum. Alan
Turing was then putting forth his idea (1936} that any explicitly stated
computational task could be performed by a machine in possession of an
appropriate finite set of instructions. He was demonstrating that, in princi-
ple, there was only one kind of computer (though, of course, there could
be many models built in many ways); and he was beginning to think about
the central issues of artificial intelligence: the relationship between human
thought and machine thought. This concern manifested itself some time
later in the famous Turing test, where a skeptic was challenged to distin-
guish the answers of a human respondent from those put forth by a
computer (see page 17). Vannevar Bush, an engineer at M.I.T., who had
suggested to the young Claude Shannon that he explore the analogy be-
tween electrical network theory and the propositional calculus, was begin-
ning to build machines that could solve differential equations. Also at this
time, as I mentioned earlier, Warren McCulloch and Walter Pitts (1943)
were developing their ideas about neural networks, specifically that any-
thing that can be exhaustively and unambiguously put into words can be
realized by a suitable finite network of neurons; thus, the brain can be
construed as a machine in a more precise way than before and, indeed, be
thought of as a Turing machine (McCorduck 1979, p. 15). And Norbert
Wiener was weaving together the strands of cybernetics, a new interdisci-
144
plinary field which investigated feedback mechanisms in organic matter
and in automata.
Finally, there was John von Neumann, in touch with all of these veins
of thought, and with perhaps the most sustained interest in the theory of
computing machines. He is generally credited with developing the idea of
a stored program, where the operations of the computer can be directed or
controlled by means of a program or a set of instructions, housed in the
computer's internal memory. (It was thus no longer necessary to reprogram
for each new task.) He demonstrated how binary logic and arithmetic
could work together in forming stored programs. One can encode instruc-
tions to the machine in the same language as that used for the data it
processes, and thus mix instructions and data in the program and store
both in the computers. These conceptual breakthroughs opened the way
for adjuncts to programming-such as assemblers, which can cull together
subroutines into the main program, and compilers, which can translate from
one language (usually a high-level programming language which is conve-
nient to use) to a more basic language, reflected in the actual electrome-
chanical operations of the computers. Finally, von Neumann pursued with
special vigor the analogies (and disanalogies) between the brain and com-
puting machines.
It is not entirely clear whether von Neumann appreciated the potential
of programs to attack and solve problems of intellectual depth. Yet he was
certainly aware of the nexus of issues in this area and if he had not died
of cancer while still relatively young, he might well have become the major
figure in the history of artificial intelligence. But this role was shared by
four scholars at the Dartmouth meeting in 1956: Herbert Simon and Allen
Newell, Marvin Minsky, and John McCarthy.
The Programs of the Dartmouth Tetrad
Programs for Problems: Allen Newell and Herbert Simon

While all the scholars at Dartmouth were actively engaged in thinking
about thinking machines, only the team of Newell and Simon had already
demonstrated that these ideas "in the air" could be implemented. Their
first program, Logic Theorist (LT), could actually prove theorems taken
from Whitehead and Russell's Principia.
Confident that their discovery marks an important point in the intel-
lectual history of science in this century, Newell and Simon have described
their progress in detail. In 1952 the two men had met at the Rand Corpora-
tion and had been impressed by the fact that the new electronic computers
were more than simply "number crunchers" and could, in fact, manipulate
145
all manners of symbols. Working with Cliff Shaw, a colleague at Rand,

Newell and Simon began to explore the kinds of symbol-manipulation
task that might be solved by a computer. Among other things, they consid-
ered the playing of chess and the solving of geometrical problems and
arrived almost as an afterthought at the proof of logical theorems.
Newell and Simon knew it would be difficult to write programs capa-
ble of complex forms of information processing directly in the language of
computers. What they needed was a "higher-level" language, one more
congenial to the human programmer, which could then automatically be
translated into the "machine language" of the computer. In 1955 the Ne-
well team began to devise such "information processing languages" (IPL)
or "list processing languages." On 15 December of that year, Simon simu-
lated "by hand" a proof from Whitehead and Russell's Principia; moreover,
the hand simulation was carried out in such detail that his colleagues
agreed that the procedure could actually be carried out on an early com-
puter called (after von Neumann) the Johnniac. Simon told his class in
mathematical modeling, "Over Christmas Allen Newell and I invented a
thinking machine" (quoted in McCorduck 1979, p. 116). And in August
1956, the Logic Theorist program actually produced on Rand's Johnniac
computer the first complete proof of a theorem (Whitehead and Russell's
theorem 2.01).
List processing was a technique developed to answer the problem of
allocating storage in a limited computer memory. Until that time, the
allocation of space had been prescribed at the beginning of a program run.
But the Logic Theorist ate up memory so rapidly and unpredictably that
users could not afford to allocate memory storage permanently. Shaw and
Newell solved the problem by labeling each area of storage, having the
machine maintain an updated list (including lists of lists) of all available
spaces, and simply making the storage space available as needed.
In addition to solving the space-allocation problem, this method of list
processing allowed the programmers to create data structures to store
information in a way that was readily accessible and that, not coinciden-
tally, may have borne a resemblance to human thought processes.
How does the computer program Logic Theorist actually work? This
program discovers proofs for theorems in symbolic logic, of the kind origi-
nally presented in Whitehead and Russell's Principia Mathematica. The pro-
gram contains the basic rules of operation-a list of axioms and previously
proved theorems. The program then receives a new logical expression and
is instructed to discover a proof. From that point on, the program runs
through all the operations of which it is capable in an effort to find a proof.
If it finds one, the proof is printed out on a long strip of paper. If not, it
declares it cannot solve the problem and ceases its operations.
146
The demonstration that the Logic Theorist could prove theorems was
itself remarkable. It actually succeeded in proving thirty-eight of the first
fifty-two theorems in chapter 2 of the Principia. About half of the proofs
were accomplished in less than a minute each; most of the remainder took
from one to five minutes; a few took fifteen to forty-five minutes; there
was a strong relation between the number of items in the logical expression
and the length of the proofs. It turned out that one proof was more elegant
than Whitehead and Russell's attempt of fifty years before-as Simon
informed Bertrand Russell, who was delighted by this ironic twist. How-
ever, the journal of Symbolic Logic declined to publish an article co-authored
by the Logic Theorist in which this proof was reported (McCorduck 1979,
p. 142).
It was still possible that this demonstration-although intriguing to
engineers or logicians-could fall outside the purview of scientists inter-
ested in the operation of the human mind. But Newell, Simon, and their
colleague Cliff Shaw stressed that they were demonstrating not merely
thinking of a generic sort but, rather, thinking of the lcind in which humans
engage. After all, Logic Theorist could in principle have worked by brute
force (like the proverbial monkey at the typewriter); but in that case, it
would have taken hundreds or even thousands of years to carry out what
it actually achieved in a few minutes. Instead, however, LT worked by
procedures that, according to the Newell team, were analogous to those
used by human problem solvers. Among the methods used by LT are
substitution of one kind of expression for another; a detachment method,
where the program works backward from something that has already been
proved to something that needs to be proved; and a syllogistic form of
reasoning, where if "a implies b" is true, and "b implies c" is true, then
'a implies c" is also true.
In a further effort to underscore parallels between human and machine
problem solving, Newell and Simon performed various experiments with
their program. They showed that, if they removed the record of previous
theorems (on which solutions to new theorems were constructed), the
Logic Theorist could not solve problems it had previously handled in ten
seconds. This was perhaps the first attempt ever to perform an experiment
with a computer to see if it "responds" in the way that human beings do.
Trying to portray their demonstration in the proper light, the Newell
team also stressed the resemblance between human and machine problem
solving. They based this claim on some protocols they had gathered on
human subjects engaged in the same tasks. In both humans and computing
machines, the team found certain staples of human problem solving. For
example, they reported certain moments of apparent insight as well as a
reliance on an executive process that coordinates the elementary opera-
147
tions of LT (for example, substitution, detaching) and selects the subprob-

lem and theorems upon which the methods operate. In conclusion, they
located their work centrally within the new cognitive vogue:
We do not believe that this functional equivalence between brains and com-
puters implies any structural equivalence at a more minute anatomic level (for
example, equivalence of neurons with circuits). Discovering what neural mech-
anisms realize these information processing functions in the human brain is a task
for another level of theory construction. Our theory is a theory of the information
processes involved in problem-solving and not a theory of neural or electronic
mechanisms for information processing. (Newell, Shaw, and Simon 1964, p. 352)
These remarks can be seen as directed at those-for instance, the McCul-

loch circle-who looked for the secret of the computer's operations (and
of thinking in general) in an understanding of how neural circuitry works.
The Simon and Newell group bet that this analogy was not helpful, and
that it would be more profitable to conceptualize problem solving at a
much more macroscopic level. For twenty-five years, this "dry" view has
carried the day: as we shall see, it is only in the last few years that an
approach that pays closer attention to what is known about the nervous
system has begun to gain support within the artificial intelligence
community.
By devising and running Logic Theorist, Newell and Simon showed
that A.l. was a possibility, if not a reality. While all claims before had, in
a sense, been handwaving, two key demonstrations had now been made:
(1) computers could engage in behavior that, if exhibited by humans,
would unambiguously be considered intelligent; (2) the steps through
which the programs pass in the course of proving theorems bear a non-
trivial resemblance to the steps observed in human problem solving.
But Newell, Simon, and their colleagues were soon after even bigger
game. Their most ambitious project was the devising of the General Prob-
lem Solver (GPS), a program whose methods (at least in principle) could
be utilized for all manner of problem solving (Newell and Simon 1972).
The General Problem Solver was capable of such apparently diverse tasks
as solving theorems, playing chess, or solving such puzzles as the mission-
ary-cannibal conundrum, the tower of Hanoi, and cryptarithmetic-a
fiendish mind bender where letters stand for numbers and the sums or
products of words yield yet other words. But GPS did not just attempt to
solve these problems in the most efficient way. Rather, it sought to mimic
the processes used by normal human subjects in tackling such problems.
Thus, a very important part of this research enterprise was the collection
of protocols that recorded the introspections and notations of subjects
engaged in problem solving.
148
ArH/icial Intelligence: The Expert Tool
The methods used by the General Problem Solver can be readily
described. In means-ends analysis, one first states the desired form of the
solution of a problem, and then compares one's present place in the process
of problem solution with the final goal desired. If these two instances
coincide, then the problem has been solved. If not, the solver (human or
mechanical) clarifies the difference and searches for methods to reduce the
difference between where one is and where one wants to go.
The art in the General Problem Solver lies in the methods of reducing
this distance. A table is set up that associates the system's goals with
operators that may be of use in achieving them. Once the difference has
been computed between the present situation and the goat the system
then selects an operator associated with that difference and tests whether
the operator is applicable to the current situation. If it can be applied, and
if it produces a result that is closer to the desired end state, it is repeated
again. If it proves inapplicable, then the system generates a subgoal, whose
aim is to reduce the difference between the current situation and the
situation where the operator can be applied. This procedure is simply
repeated until the goal is achieved or it has been demonstrated that it
cannot be achieved with the information given, or with the operators
available in the program.
The General Problem Solver also exhibited other features designed to
facilitate problem solving. It was possible to decompose the program into
subproblems which could be tackled one at a time. It was possible to ignore
some of the complicating factors in a situation in order to arrive at a plan
of attack. It was possible to omit certain details of a problem as well. For
example, in solving a problem in the propositional calculus, the machine
can decide to ignore differences among the logical connectives and the
order of symbols, only taking into account what the symbols are and how
they have been grouped.
While the General Problem Solver was eventually abandoned because
its generality was not as great as its creators had wished, and because the
field of artificial intelligence moved in different directions, the program can
be regarded as the first to simulate a spectrum of human symbolic behav-
ior. GPS also occupied a major role in Simon and Newell's thinking about
the enterprise in which they were engaged. As they conceived it, all intelli-
gence involves the use and manipulation of various symbol systems, such
as those featured in mathematics or logic. In the past, such manipulation
had been done only by the human being within the confines of his own
head, or with paper and pencil, but, with the advent of the digital com-
puter, symbol manipulation has become the province of electronic machin-
ery as well. On the Newell-Simon account, the computer is a physical
symbol system like the human brain and exhibits many of the same prop-
149
II I THE CoGNITIVE SciENCES : HisToRICAL PERSPECTIVE
erties as the human being whose behavior it has now been programmed
to simulate.
Just as the cell doctrine has proved central in biology, and germ theory
is pivotal in the area of disease, so the concept of physical symbol system is
deemed by Simon and Newell and their colleagues at Carnegie-Mellon as
the core doctrine in the area of computer science. Proceeding in Boolean
spirit, the job of the theorist is to identify that set of processes which
operates on symbolic expressions in order to produce other expressions
that create, modify, reproduce, and/or transform symbolic structures. A
physical symbol system is necessary and sufficient to carry out intelligent
actions; and, conversely, any system that exhibits general intelligence will
prove upon analysis to be a physical symbol system. Such a system con-
sists of a control, a memory, a set of operations, and input and output: its
input consists of objects in certain locations; its processes are operations
upon the input; its output is the modification or re-creation of objects in
certain locations.
A key notion in the Newell-Simon scheme is the production system,
in which an operation will be carried out if a certain specific condition is
met. Programs consist of long sequences of such production systems oper-
ating on the data base. As described by the theorists, the production
system is kind of a computational stimulus-response link; so long as the
stimuli (or conditions) are appropriate, the response (or production) will
be executed. In the course of developing the General Problem Solver,
Simon and Newell had propounded a perspective on artificial intelligence,
a theory of thinking, and an agenda for future research.
The vision of Newell and Simon was formidable. From their perspec-
tive, the profound similarities between the human mind engaged in solving
a problem, and the computer programmed to solve the same problem, far
overrode differences in hardware (an electronic machine versus a parcel of
neural tissue). Both are simply systems that process information over time,
proceeding in a more or less logical order. Moreover, to the extent that the
steps noted by an introspecting individual paralleled the lines of a com-
puter program, one was no longer simply engaging in weak A.I.: it made
sense to think of this man-made physical symbol system as actually
engaging in problem solving.
Critics of the Newell-Simon effort brought up a number of issues.
First of all, all the information in the computer program had been placed
inside the program by humans: thus, to put it colloquially, the problem
solver was only doing what it was programmed to do. For instance, it
was Newell and Simon who structured the problems given to the pro-
gram and, in some cases, determined the order in which they were pre-
sented. Use of terms like insight was but a misleading metaphor. To
150
Newell and Simon, this criticism appears anachronistic: so long as the
program was not simply engaging in rote repetition of steps but actually
used rules to solve problems to which it had not previously been ex-
posed, its behavior was as "intelligent" as that of a human being. Simply
old-fashioned habits of thought induced critics to withhold the term
intelligent. Indeed, it was necessary to think of humans as being pro-
grammed with rules, just like computers. Not all scholars were convinced
by this "democratic response."
Another line of criticism centered on certain differences between
human beings and computer programs. For example, human beings can
improvise shortcuts or heuristics, whereas computers will repeat the same
processes unless they can be programmed to learn from earlier efforts.
Recognizing this limitation in the General Problem Solver, Newell and
Simon set out to devise programs capable of learning.
A final line of criticism involves the kinds of problem posed to the
General Problem Solver. Despite its ambitious name, the problems were all
puzzles or logical challenges: these lend themselves to expression in sym-
bolic forms, which in turn can be operated upon. This restriction to
"closed" questions was essential since GPS could tackle only mathemati-
cal-logical problems. Clearly, many problems confronted by humans (such
as finding one's way about a forest or learning to master a dance) are not
readily expressed in symbolic logic. Here we encounter a revealing instance
of how notions of "thinking" or "problem solving" may be artificially
constrained by the programs that currently exist.
Some of these criticisms pertain to other efforts in artificial intelligence
as well, but each separate line of inquiry deserves consideration on its own
merits. Let me therefore turn more briefly to what other principal inves-
tigators were doing in the first decade or so following the Dartmouth
conference.
Marvin Minsky and His Students

Marvin Minsky at M.I.T. has not been as active a contributor to the
published literature as Newell and Simon, nor is a single line of work
particularly associated with his own laboratory. But as a seminal thinker
about artificial intelligence, who had arrived independently at some of
Newell and Simon's ideas, and as a mentor of an active cadre of talented
students, he played a significant role in the progress wrought by artificial
intelligence in the 1960s (and thereafter).
Under his inspiration, Minsky's students have led artificial intelli-
gence in directions other than those explored by Newell and Simon. One
student, T. G. Evans, devised a program in the late 1960s which solved
151
I I I THE CoGNITIVE SciENCES : HisToRICAL PERSPECTIVE
analogies of a visual sort (1968). Expressed anthropomorphically, the pro-

gram was shown a pair of figures that bore some relation to one another,
and was asked to select another set of figures from a set that completed
the visual analogy. Thus, for example, the program is shown "A is to B"
and must then pick that picture out of five which indicates the relations
that obtain between "C" and "D."
It must be noted that the program does not solve the visual analogy
by use of perceptual "pick-up" mechanisms (of the sort humans might
use), but converts the description into symbolic forms of the sort that
would be used in a numerical analogy problem. The program accomplishes
the analogy by describing both A and B as figures, and then characterizing
the difference between the descriptions (in terms like inside, above, left of,
rotated, or reflected); next, it applies the identified difference as a transforma-
tion rule to C, in order to arrive at a pattern having the same description
as one of the five candidate numbered patterns (Boden 1977). Evans's
program performs this task at the level of a high school sophomore. To
program at that level, Evans had to build one of the most complex pro-
grams that had ever been written. The machine had about a million bits
of memory, and the program had to use every bit of it.
Another individual working in Minsky's laboratory, Daniel Bobrow,
adapted the work on problem solving to a linguistic domain (1968). Bo-
brow's STUDENT program was designed to solve the kinds of algebra
problem which youngsters encounter in high school mathematics books.
As an example, one of the problems posed in Bobrow's thesis went like
this:
The gas consumption of my car is 15 miles per gallon. The distance between
Boston and New York is 250 miles. What is the number of gallons of gas used on
a trip between New York and Boston? (Bernstein 1981, p. 113)
As described by Marvin Minsky, the program assumed that every

sentence is an equation; it was given knowledge about certain words to
help it locate the equation. For example, the word is was coded to mean
equal amounts, on both sides of an equation. Per meant division. The
program was driven by these desired meanings to analyze the syntax. In
Minsky's words:
From the mathematical word "per" in that first sentence's "miles per gallon,"
it can tell that the number 15 would be obtained by dividing a certain number
x of miles, by some other number, y, of gallons. Other than that, it hasn't the
slightest idea what miles or gallons are, or, for that matter, what cars are. The
second sentence appears to say that something else equals two hundred and fifty
miles-hence the phrase "the distance between" is a good candidate to be x. The
152
third sentence asks something about a number of gallons-so that phrase of "gas
to be used on a trip" is a candidate to bey. So it proposes one equation: x = 250,
and another equation, xly = 15. Then the mathematical part of the program can
easily find that y = 250/15. (Bernstein 1981, p. 113)
STUDENT illustrates well the powers that were being exhibited by

programs in the mid-1960s, as well as certain limitations for which they
were criticized. That the programmers were extremely clever, and that the
machines could often carry out feats that, if executed by human beings,
would unquestionably be considered intelligent, was difficult to dispute
(though certainly some critics did [see Arnheim 1969; Dreyfus 1972]). And
yet it seemed equally clear that the procedures used were often completely
at variance with those ordinarily used by ordinary people. Faced with the
preceding problem, a flesh-and-blood student would think about the na-
ture of an automobile trip, the geographic locations of New York and
Boston, and what happens when you are using up gas on such a trip.
Indeed, the student would almost have to look through these particulari-
ties in order to figure out just what the actual mileage per gallon should
be (even as he might hazard a plausible guess based simply on his own
"real world" experience).
In the case of the computer, however, the procedure was almost the
exact opposite. The computer had no idea what the problem was about,
and would have performed in exactly the same way had the issue been
pennies for peanuts or millions for missiles. The computer's knowledge
was purely syntactic. The program is designed to expect certain statements
about equalities and to draw the most probable inference about which
entities in the problem are likely to constitute the principal components
of an equation. It becomes a relatively trivial matter to confuse the program
-for example, by including the word is in a context where it does not
denote an equation but rather forms part of a relative clause or an inciden-
tal remark. Similarly, one can wreak havoc by injecting the word per as part
of an expression, such as per capita or per chance. The human subject will
sometimes be fooled by some extraneous fact which looks relevant (for
example, the cost of the gas). The computer's difficulty is that it cannot
look through the particular way in which it has been programmed in order
to pick up the actual reference of a word or a number. Having no insight
about the subject matter of a problem, the computer is consigned to make
blunders that, in human beings, would never happen or would be consid-
ered extremely stupid.
Indeed, the computer resembles a human being who is asked to solve
an algebra problem in a foreign language, of which one knows but a few
words: faced with such an enigma, both computer and foreigner gravitate
153
to the few numbers each recognizes, and make the best guess about the
mathematical operations that should be carried out on those numbers.
While the problems selected by Minsky and students span a wider gamut
than those tackled by GPS, they are always reformulated as symbolic
expressions of a canonical sort. Nonetheless, it must be stressed that, as
with the Newell-Simon efforts, the programs worked. The kinds of exer-
cise expressed in ordinary language, over which schoolchildren have
struggled for generations, could be solved by a mechanical process.
Lists and Logics: fohn McCarthy

While the Simon and Minsky laboratories were busily engaged in
fashioning demonstration programs, John McCarthy-first at M.I.T., later
at Stanford-was engaged in less flashy but equally important endeavors.
One of his major accomplishments was to design LISP (for "list process-
ing"), the computer language that became most widely used in the field of
artificial intelligence (McCarthy et al. 1962; Foster 1967).
As I have indicated with respect to Newell and Simon's early work,
it was important for workers to have a language in which they could think
readily about problem solving, and one that mimicked closely the kinds
of mental step through which a human problem solver putatively passes.
LISP and LISP-like "higher-order" languages came to be considered the
mathematics of artificial intelligence, the precise and unambiguous argot
in which theories are cast (see also Boden 1977 and Winston 1977). The
language is basically concerned with the presentation and manipulation of
lists, of items on lists, and of lists of lists, each of which can be named.
Both programs and data are structured as lists. LISP's power derives from
the fact that it is a recursive programming language, one well suited to the
description and manipulation of structures and sets of structures. As a
recursive language, it is hierarchical and can be described (and can operate)
at several levels of detail. LISP is also very flexible: the program can move
among levels that are nested within one another, can refer to and operate
upon itself as often as necessary, and can automatically reallocate bits of
memory. For these and other reasons, LISP and its descendants have con-
tinued to be used by most cognitively oriented workers in the computer
sciences.
But McCarthy is more than a mere inventor of a useful language. He
also has had strong ideas about the goals of artificial intelligence and how
they can best be achieved. McCarthy believes that the route to making
machines intelligent is through a rigorous formal approach in which the
acts that make up intelligence are reduced to a set of logical relationships
or axioms that can be expressed precisely in mathematical terms. With
154
Patrick Hayes, then of Edinburgh University, McCarthy wrote a seminal
article in 1969, in which he argued for the use of a formal predicate calculus
substrate embedded within a system designed for understanding language
(McCarthy and Hayes 1969). He called for the formalization of concepts
like causality, ability, and knowledge. If such an approach were to be adopted,
it would prove possible to use theorem-proving techniques that are not
dependent on the details of particular domains. McCarthy's system was
based upon a faith in the consistency of a belief system and upon a view
that all knowledge is (or can be) thought of in purely logical terms. As we
shall see, these assumptions have had relatively few proponents in recent
years. Still, McCarthy has adhered to the general program. He has been
designing a nonconventional modification of standard logic in order to
model common-sense reasoning, and his former associate Hayes (1982) has
been trying to formulate in logical terms the thinking processes of "naive
physics"-the physics of the man in the street. McCarthy stands as an
extreme proponent of one point of view in artificial intelligence-a man
who holds high standards for the field and is less willing than others to
bend to the practical or to the particular.
Other Programming Milestones
While mainstream scholars in the 1960s were either pursuing logical prob-
lem solving or theorizing about the proper course of artificial intelligence,
certain other work undertaken during that era held considerable implica-
tions for the future. One such line was carried out by Edward Feigenbaum,
an early student of Simon's. In collaboration with Joshua Lederberg, a
Nobel Laureate in genetics at Stanford, Feigenbaum decided to analyze
data from a mass spectrograph (Feigenbaum, Buchanan, and Lederberg
1971). The result was a program named DENDRAL which was designed
to figure out (on the basis of a vast amount of data from mass spectro-
graphs) which organic compound was being analyzed. After processing the
relevant data, the program formulated hypotheses about the compound's
molecular structure and then tested those hypotheses by way of further
predictions. The ultimate output was a list of possible molecular com-
pounds ranked in terms of decreasing plausibility. DENDRAL's perform-
ance compared favorably with that of expert chemists (Boden 1977).
As a program, DENDRAL departed from Simon and Newell's classical
model in two ways. First of all, DENDRAL made use of considerable stored
knowledge about chemicals. It was predicated on the assumption that
experts have a great deal of knowledge and as much of it as possible ought
to be possessed by any program that purports to behave intelligently.
155
Second, if seemingly at odds with this assumption, was the decision not
to try to emulate the ways in which human beings would solve the prob-
lem. Feigenbaum was explicitly interested in top-flight performance, and
not particularly intent on mimicking the way that a human being would
operate when confronted with the same closetful of information. Here we
see an early sign of the sharp cleavage that was to ensue between simula-
tion of human behavior and cultivation of intelligent performances by
whatever means.
Two other investigative lines in the 1960s attempted to enter into a
seemingly elusive area of human performance: meaningful conversation
between individuals. In view of Turing's famous challenge (see pages
17-18), this was a tempting goal to pursue. Kenneth Colby, a psychiatrist
then working at Stanford, and Joseph Weizenbaum of M.I.T. both wrote
programs that modeled two-person dialogues. In Colby's case, he was
interested in psychiatric disorders and so produced a program whose an-
swers might be given by a paranoid personality with a particular obsession
about the Mafia:
Why are you in the hospital?

I SHOULDN'T BE HERE.
Who brought you here?
THE POLICE.
What trouble did you have with the police?
COPS DON'T DO THEIR JOB.
What do they do instead?
COPS ARREST THE WRONG PEOPLE.
Are you Italian?
MY BACKGROUND IS BRITISH BUT I WAS BORN IN THIS COUNTRY.
Do you know any Italian crooks?
I TRY TO AVOID THE UNDERWORLD.
(Boden 1977, pp. 100-1; see also Colby 1975)
Weizenbaum sought to reproduce the behavior of a Rogerian thera-

pist, who assumes a nondirective stance toward clients:
Men are all alike.

IN WHAT WAY?
They're always bugging us about something or other.
CAN YOU THINK OF A SPECIFIC EXAMPLE?
Well, my boyfriend made me come here.
YOUR BOYFRIEND MADE YOU COME HERE.
He says I'm depressed much of the time.
I AM SORRY TO HEAR THAT YOU ARE DEPRESSED.
It's true. I am unhappy.
156
DO YOU THINK COMING HERE WILL HELP YOU NOT TO BE UN-
HAPPY?
I need some help, that much seems certain.
WHAT WOULD IT MEAN TO YOU IF YOU GOT SOME HELP?
Perhaps I could learn to get along with my mother.
TELL ME MORE ABOUT YOUR FAMILY.
(Boden 1977, pp. 106-7; see also Weizenbaum 1966)
Both of the programs were disguised with sufficient artfulness that

they might, for a time, fool an unsuspecting person conversing with the
computer or a naive witness of an interchange. Yet both devisers concede
that anyone who knew the design of the program, or was able to make the
correct inference, could readily unmask (or "defrock") the program. The
reason is that, as in the case of Bobrow's STUDENT program, the programs
do not in any interesting sense understand the words being used but are
simply constructed in such a way that certain key words will trigger fixed
responses. For example, Weizenbaum's ELIZA depends on pattern-match-
ing techniques. Each sentence in the input text is scanned for key words
like L you, like, and father: if one of these words is found, the sentence is
transformed according to a rule associated with that word. In contrast, if
no such key word is found in the input text, ELIZA responds with a
content-free formula, such as "WHY DO YOU THINK THAT?" or with
a neutral referral to some earlier remark. There are some other fiendish
features. For example, suppose that in the absence of a recognizable key
phrase, a decision is made to revert to an earlier topic: the program is so
devised that it will return to a topic that had been prefixed by the term
mine on the not unreasonable assumption that such topics are likely to be
"charged" for the conversant.
It should be noted, not entirely parenthetically, that while the Colby
and the Weizenbaum research programs were developed in similar fash-
ions, and indeed initially involved some collaboration, the two men have
evolved diametrically opposed attitudes toward artificial intelligence. They
also have engaged in a heated personal dispute. Weizenbaum, it seems,
became disaffected with the whole field of artificial intelligence: he was
appalled that some individuals in the clinical professions and in the mass
media took his rather whimsical demonstrations as a serious therapeutic
tool which might actually be used with disturbed human beings. Colby,
on the other hand, is a "true believer," who feels that computers have an
important role to play in research about mental illness and possibly in its
treatment as well. The issues between these two individuals run deep and
touch on the humanistic aspect of the use of computers-a theme that has
come up repeatedly in recent debates about the value, and the values, of
artificial intelligence.
157
I I I THE CoGNITIVE SciENcEs : HISTORICAL PERSPECTIVE
The Phenomenal SHRDLU
While the Weizenbaum and Colby programs make no pretense of actually

understanding the utterances to which they are responding (being
"dumbly" keyed instead to certain triggering phrases), a different story
concerns a program devised around 1970 by an M.I.T. doctoral student,
Terry Winograd. Called SHRDLU (after the seventh through the twelfth
most frequent letters in a printer's array), this program operates in a way
remote from those I have so far described. Rather than being a general
problem solver, or a responding mechanism simply activated by key
words, SHRDLU purports to be an understanding expert-albeit one
working in a very limited domain. Winograd designed for his program a
make-believe world of simple blocks which could (if they really existed)
be piled and stacked in various ways: his program is sufficiently sophis-
ticated in its linguistic knowledge that it can carry out a complex set of
instructions. Moreover, SHRDLU issues signs that these instructions are,
in fact, understood-one of the clearest signs of that understanding being
that the program asks for clarification in cases where instructions are
ambiguous.
What does SHRDLU do? Briefly, it replies to questions about its block
world as posed by an interviewer, executes commands on the simulated
objects in the block world, and reports the results. Sometimes these are
simple actions like grasp, move, or put; while at other times, they are actions
at a higher level of abstraction, like slack and clear lop. Before slack can be
handled, for example, the tops of both objects that will form the stack must
be cleared. The program searches for actions that will achieve each subgoal
and for sets of preconditions for each action in turn. This task may involve
the construction of a sequence of actions from the goal state back to the
state in which all the preconditions of the action necessary to initiate the
chain have been satisfied.
Let us eavesdrop on some sample snatches of SHRDLU's expertise:
Pick up a big red block.

OK
Find a block which is taller than one you are holding and put it into the box.
OK
What does the box contain?
THE BLUE PYRAMID AND THE BLUE BOX.
158
What is the pyramid supported by?
THE BOX
Can a pyramid be supported by a block?

YES
Will you please stack up both of the red blocks and either a green cube or a
pyramid?
OK
Which cube is sitting on the table?
THE LARGE GREEN ONE WHICH SUPPORTS THE RED PYRAMID.
Is there a large block behind a pyramid?
YES, THREE OF THEM: A LARGE RED ONE, A LARGE GREEN CUBE
AND THE BLUE ONE.
When did you pick up (the green pyramid)?

WHILE I WAS STACKING UP THE RED CUBE, A LARGE RED BLOCK,
AND A LARGE GREEN CUBE.
Why?
TO GET RID OF IT.
Why did you do that?
TO CLEAN OFF THE RED CUBE.
Put the blue pyramid on the block in the box.

OK
Thank you.
YOU'RE WELCOME.
(Winograd 1972, quoted in Boden 1977, pp. 115-22)
Compared with the level of understanding exhibited by other pro-

grams, SHRDLU was a major breakthrough. To be sure, its simulated
world was small, and the number of actions it could carry out and of
questions it could answer, extremely limited. Still, within its particular
universe, the program performed in a plausible way, "perceiving" distinc-
tions and "acting out" commands, as if to suggest that it actually under-
stood what was being asked of it. (Whether it can be said to really under-
stand is a question of tremendous controversy, as we shall shortly see.)
Winograd's program was more sophisticated than its predecessors
owing to its use of a series of expert parsers or specialists: a syntactic
specialist which segments the sentence into meaningful word groups; a
sentence specialist which determines how the objects described by a noun
group relate to the action described by a verb group; and a scenario special-
ist which understands how individual scenes relate to one another and to
the sequential story they must collectively tell. For example, in the block
world situation, the scenario specialist must handle pronoun references (no
159
II ·I THE CoGNITIVE SciENCES : HisTORICAL PERSPECTIVE
mean task) and the temporal sequence of all actions performed. In addition
to these experts which can flexibly interact with one another and share
information, the program also has belief sys,tems, knowledge about prob-
lem solving (such as the mechanisms of deduction), and specialists that
detect whether an utterance is question, command, or comment.
SHRDLU was, for its time, a stunning demonstration of apparent
language production and understanding capacities. Yet, according to Dan-
iel Dennett (1978), one of its chief contributions lay in another sphere.
Specifically, the SHRDLU experiment explored some of the extensive de-
mands imposed on any system that undertakes to follow directions, plan
changes in the world, and then keep track of them. While the ways in
which these tasks are done may not simulate the way that a real person
would work in a real block world, the procedures devised by Winograd
were ingenious and at least suggested the kinds of problem that would
have to be confronted by any system seeking truly to understand, rather
than simply to mimic understanding.
Certain limitations should be noted, however. SHRDLU does not have
adequate semantic information to appreciate the differences in meaning
between function words like and, the, and but. More tellingly, SHRDLU
lacks any ability to leam to perform better. Its knowledge suffices for it to
know why it does what it does but not enough for it to remember what
has gone wrong when a failure occurs, or to learn from error to make more
appropriate responses in the future. A later block-world program, designed
by Gerald Sussman (1975) and called HACKER, showed that such learning
was, in fact, possible.
Winograd's program came into being at a crucial moment in the his-
tory of artificial intelligence. There were furious debates going on about
that time, both within the artificial intelligence community and between
it and its severest critics. In many ways, the Winograd program spoke to
these debates, even if it did not singlehandedly resolve them.
Pivotal Issues
The Need for Expert Systems

First, debates within the A.I. community. The 1960s were a time of
excitement about general problem solving. Led by the redoubtable Newell
and Simon, the search was on for programs that could, at least in principle,
deal with every manner of material. But by the late 1960s, the limitations
of these programs, which had to be couched in highly general terms, were
becoming increasingly evident.
About this time, Edward Feigenbaum returned to Carnegie, his alma
160
mater, to deliver a talk to an audience that included Newell, Simon, and
other leading cognitive scientists. Feigenbaum threw out a challenge to his
former teachers: "You people are working on toy problems. Chess and
logic are toy problems. If you solve them, you'll have solved toy problems.
And that's all you'll have done. Get out into the real world and solve
real-world problems" (Feigenbaum and McCorduck 1983, p. 63). Feigen-
baum had brought along information about DENDRAL, his expert system
which incorporated massive amounts of specific knowledge about organic
compounds. While DENDRAL was achieving impressive successes, it ini-
tially met a skeptical audience. As Joel Moses of M.I.T. has commented:
The word you look for and you hardly ever see in the early AI literature is
the word knowledge. They didn't believe you have to know anything, you could
always rework it all .... In fact 1967 is the turning point in my mind when there
was enough feeling that the old ideas of general principles had to go. . . . I came
up with an argument for what I called the primacy of expertise, and at the time
I called the other guys the generalists. (Quoted in McCorduck 1979, pp. 228-29)
Moses noted Allen Newell's intense objections to his approach:
He called my position the big-switch theory, the idea being that you have all
these experts working for you and when you have a problem, you decide which
expert to call in to solve the problem. That's not AI, you see.... I think what finally
broke Newell's position was Winograd. (Quoted in McCorduck 1979, p. 229)
And, indeed, shortly after Winograd's program had been completed, even
the leaders of the old guard, like Marvin Minsky and Allan Newell, be-
came convinced of the limitations of generalist programs and of the need
for systems possessing considerable specialized or expert knowledge. Yet
analogous issues were to return in other guises: for example, the question
whether there should be a general language for all artificial intelligence
programs or many specifically crafted tongues.
Procedural versus Declarative Representation

As for the preferred manner of programming, there was at the same
time a vigorous battle between scholars who favored declarative representa-
tion-knowledge coded essentially as a set of stored facts or declarations
-and those who favored procedural representation-knowledge coded as a
set of procedures or actions to be carried out. In the early 1970s, the
respective camps were sharply divided, with those wedded to LISP typi-
cally veering in favor of declarative knowledge. In their view, declarative
languages are easy for people to understand and use. Moreover, such
programs are economical, since a bit of information has to be stored only
161
once and can then be tapped at will. Declarativists felt that intelligence
rests on a highly general set of procedures which can be used widely,
coupled with a set of specific facts useful for describing particular
knowledge domains; they were not convinced that knowledge of a sub-
ject matter is necessarily bound up with the procedures entailed in its use
in one or another context.
In a manner reminiscent of the functionalists in psychology, the
proceduralists felt that human intelligence is best thought of as a set of
activities that individuals know how to do; whatever knowledge is neces-
sary can be embedded in the actual procedures for accomplishing things.
Many things that we do know how to do are best viewed as procedures;
and, indeed, it is difficult to describe them formally in a declarative way.
For example, if one wants to build a robot to manipulate a simple world,
one does it most naturally by describing these manipulations as procedural
programs. Procedural representation has the additional advantages of al-
lowing ready use of higher-order (second-level) control information or
knowledge of where one routine in a program has to be triggered by
another, and being applicable across several domains (Boden 1977; Cohen
1977; Newell 1983; Winograd 1975).
The debate about procedural versus declarative representation has
become somewhat muted in the last several years. There is now increasing
recognition that not all computing functions lend themselves better to one
mode of representation than to another; and that, in fact, some problems
are better handled by one approach, other problems by its rival. Initially
Winograd had been a chief proponent of procedural systems; but he him-
self wrote a paper describing the advantages of these two different modes
and, more recently, has collaborated with Daniel Bobrow in devising sev-
eral Knowledge Representational Languages which incorporate both
procedural and declarative components (Bobrow and Winograd 1977).
The Three Sharpest Cuts

Also in the early 1970s, at the same time that these intramural debates
were going on among scholars generally in sympathy with the goals of
artificial intelligence and at odds only about the optimal means, far more
critical examinations of artificial intelligence were also afoot. Joseph Wei-
zenbaum (1976), an early practitioner who had devised the seductive
ELIZA, launched a strong attack on A.I. enthusiasts. In his view, many of
their claims were excessive, wholly out of line with what had actually been
achieved. Moreover, he was critical of the future aspirations of artificial
intelligence; he believed that many of the tasks now being assigned to
machines had best be left to human beings ("Render unto Johnniac ... "),
162
Artificial Intelligence: 77ze Expert Tool
and that a dangerous confusion was taking place regarding what was
properly the realm of humans and what ought to be ceded to machines.
People are entities that are wholly different from machines; such uniquely
human experiences as love and morality must remain sacrosanct.
In an even more wide-ranging attack on artificial intelligence, Hubert
Dreyfus, a phenomenologically oriented philosopher at the University of
California in Berkeley, had published in 1972 a critical book called What
Computers Can't Do: A Critique of Artificial Reason. Dreyfus made much in this
book of the fundamental differences between human beings and comput-
ers: he claimed that, unlike computers, human beings have a fringe con-
sciousness; a tolerance for ambiguity; a body that organizes and unifies
one's experience of objects and subjective impressions; the potential for
boredom, fatigue, or loss of drive; and clear purposes and needs that
organize one's situation.
According to the tale Dreyfus spins, after an initial run of apparent
successes, artificial intelligence had bogged down because it had no way
of coming to grips with these fundamental differences between human
beings and machines. It was wedded to the notion that all human behavior
-including all intelligence-can be formally described by logical rules.
But human life is only as orderly as necessary; it is never completely
rule-governed: life is what humans make it and nothing else. Since a
computer is not involved ("engaged") in a situation, since it has no needs
or wants or purposes, it must treat all facts as equally relevant at all times:
it cannot make the kinds of discrimination and evaluation that are the stuff
of human life and make it meaningful.
Weizenbaum's book was greeted critically, but with a modicum of
respect, by his colleagues in artificial intelligence; on the other hand, hardly
any computer scientist had a good word to say about Dreyfus's harsh
verdict. Clashes of personality and even charges of intellectual incompe-
tence came to dominate the debate. Regrettably, there was little serious
discussion between critics and enthusiasts about the issues raised-possi-
bly because of the different value systems involved. If one believes (with
Weizenbaum) that there are certain areas where computers should not be
used, this is an ethical, or perhaps a religious, judgment; and if one believes
(with Dreyfus) in a phenomenological approach to understanding, where
the feelings of the experiencing human body are central, one is committed
to an epistemological tradition foreign to virtually everyone in the world
of computer science and artificial intelligence.
(I ought to point out that Dreyfus's book went into a second edition
in 1979, and that he feels that certain trends in artificial intelligence-for
example, the adoption of organizing schemas or frames-go some distance
toward incorporating the human approach to experience. As for the com-
163
puter science community, at least a minority now feels that Dreyfus has
raised issues that deserve to be addressed seriously. Nonetheless, Dreyfus's
next book is tentatively entitled Putting Computers in Their Place. )
Perhaps most disturbing to A.I. workers was the negative review given
to their field by a supposedly disinterested English observer, Sir James
Lighthill, who had been requested by his government's Science Research
Council to evaluate the state of the art in British artificial intelligence.
Lighthill found relatively little to admire and wrote disparagingly about
the distance between initial expectations and actual achievements in the
first twenty years:
Most workers in AI research and in related fields confess to a pronounced

feeling of disappointment in what has been achieved in the last 25 years. Workers
entered the field around 1950, and even around 1960, with high hopes that are very
far from being realized in 1972. In no part of the field have the discoveries made
so far produced the major impact that was then promised.... In the meantime,
claims and predictions regarding the potential results of AI research had been
publicized which went even farther than the expectations of the majority of work-
ers in the field, whose embarrassments have been added to by the lamentable
failure of such inflated predictions. (1972, p. 17; see also Lighthill et al. 1973)
Lighthill went on to comment:
When able and respected scientists write in letters to the present author that
AI, the major goal of computing science, represents "another step in the general
process of evolution"; that possibilities in the 1980s include an all-purpose intelli-
gence on a human-scale knowledge base; that awe-inspiring possibilities suggest
themselves based on machine intelligence exceeding human intelligence by the
year 2000 [one has the right to be skeptical]. (1972, p. 17}
There was one bright note in Lighthill's deflationary document. He

singled out for special approval Winograd's thesis: in his view, SHRDLU
succeeded by using principles that suggest genuine knowledge and sensi-
tivity to the demands of natural language within a limited universe of
discourse. That Lighthill was able to make this distinction indicates that
he was not bent on dismissing all of artificial intelligence equally, and also
points once again to the special contribution made by this one dissertation
which appeared in the early 1970s.
Indeed, it seems fair to say that artificial intelligence has bounced back
from these knocks of the early 1970s and has had some singular successes
during the late 1970s and early 1980s. As Margaret Boden, a philosopher
sympathetic to A.I., has declared:
Suffice it to say that programs already exist that can do things-or, at the very
least, appear to be beginning to do things-which ill-informed critics have asserted
164
a priori to be impossible. Examples include: perceiving in a holistic as opposed to
an atomistic way; using language creatively; translating sensibly from one language
to another by way of a language-neutral semantic representation; planning acts in
a broad and sketchy fashion, the details being decided only in execution; distin-
guishing between different species of emotional reaction according to the psycho-
logical context of the subject. (Boden 1981, p. 33)
Even as the sights have been lowered from the u absurd overoptimism" of
the 1950s, the actual achievements by certain workers are notable and have
convinced most observers outside of the field itself that the experiments
are at least interesting and should be taken seriously.
Innovations in the 1970s
While the critics of A.I. were not easily silenced, there were, by general
consensus, a new burst of energy and several significant achievements in
the field in the early 1970s. This second wave is epitomized by Winograd's
SHRDLU, the shift from generalist to expert knowledge systems, and the
fusion of features of declarative and procedural approaches. At this time,
there was another vital, though rather controversial trend: the increasing
use of a top-down approach to the understanding of language and other
cognitive domains.
As exemplified in the work of Roger Schank and his colleagues at Yale
University (Schank 1972; Schank and Abelson 1977), an "understanding"
mechanism has several expectations of what a text is like in general; it also
incorporates a core set of knowledge structures about details of the subject
matter under discussion. These structures are built in as part of a prior
knowledge base; they can then be brought to bear upon a particular text
in an effort to comprehend how that text resembles, but also how it differs
from, other instances of its genre. In the best-known formulation, Schank
has introduced the notion of a script-the canonical set of events one can
expect in an often encountered setting, such as a meal at a restaurant or
a visit to a doctor's office. The script then allows one to make sense of
different meals, ranging from a snack at McDonald's to a banquet at
Maxim's; or of a series of visits to different medical specialists. Such a
structured framework allows the "understander" to deal expeditiously
with a variety of otherwise difficult-to-assimilate texts (much as Frederic
Bartlett's story schemas allowed his subjects to make sense of an otherwise
mysterious ghost story).
Another influential top-down approach is Marvin Minsky's (1975)
165
II I THE CoGNITivE SciENcEs : HISTORICAL PERSPECTIVE
notion of a frame: an expected structure of knowledge about a domain that

consists of a core and a set of slots. Each slot corresponds to some aspect
of the domain being modeled by the frame. In a frame, a description is
created and then maintained by substituting observed for predicted values.
Thus, for example, when a robot enters a room through a doorway, it
activates a "room frame" which leads into working memory and arouses
a number of expectations about what might be seen next in a typical room.
If the robot then perceives a rectangular form, this form, in the context of
a room frame, might suggest a window. The slots include those at a top
level-fixed parameters representing things that are always true about a
proposed situation (like a room having four walls). Lower levels have
many terminals-slots that must be filled with specific instances of data:
for example, objects, like a window, that one is more or less likely to
encounter in a room. It is assumed that individuals possess many hundreds
of organizing and interpreting frames, and that combinations of these
frames will be invoked in any reasonably complex situation.
Pluralisms
Minsky has also put forth an intriguing conception about how the
mind works, leading to novel proposals about how. computer programs
should be crafted. Rather than believing in a simple general processor or
central processor, through which all information must be passed, or in an
organized or unified mind that oversees all activity, Minsky now views
mental activity as consisting of many agents, each of which is a specialist
of some sort. According to this "society of minds" view, the mind consists
of several dozen processing centers, or "agents," which can handle differ-
ent types of knowledge simultaneously. Each of the local agents has a
function, which is called upon in certain circumstances, and each has access
to other agents. The agents communicate by emitting excitation and inhi-
bition rather than by transmitting symbolic expressions. They can also
censor information, much like a Freudian superego. Under this scheme,
some parts of the mind know certain things, while other parts know things
about the former. In fact, knowledge about which agents can know or
accomplish which things becomes a crucial component of this new way of
thinking about the mind (Minsky 1979, 1982).
Minsky's idea of a frame and his "society of minds" metaphor are not
in themselves theories which can be subjected to clear-cut scientific tests,
but are better thought of as organizing frameworks (frames, if you will)
which lead to the devising of programs that perform more effectively and
model human behavioral activity more faithfully. In that sense, his ideas
can be seen as reactions: reactions against those approaches that fail to
166
Artiftdal Intelligence: The Expert Tool
build in prior knowledge or expectations, and as well against those ap-
proaches that feature detailed knowledge of a specific area but have abso-
lutely no generality or connection to any other domain. The ultimate
impact of Minsky's new ideas on artificial intelligence is not yet known.
With these widespread shifts in how the knowledge base is conceptu-
alized, there has also been increasing dissatisfaction with the kind of
computer that usually serves the A.l. researcher: the serial digital "von
Neumann" computer. Scholars like Minsky himself, and younger ones like
Geoffrey Hinton and James Anderson (1981), raise an intriguing possibil-
ity: since the brain is itself a parallel rather than a serial mechanism, with
millions of neural events occurring simultaneously, the simulation of
human activities ought to be carried out by computers that also operate in
parallel fashion. The brain is an apparatus that learns, that executes many
special-purpose activities, and that has information dispersed throughout
large reverberating circuits. Hinton, Anderson, and their colleagues call for
a computer that more closely parallels the operation of the human brain,
and for programs that feature many cooperating individual agents. Most
of these efforts thus far involve the simulation of visual processing-an
area well enough understood in neurophysiological terms to permit a plau-
sible anatomical simulation. A growing number of experts speculate that
the next innovative wave of artificial intelligence will utilize "paraHel
kinds" of architecture for information processing (Feigenbaum and
McCorduck 1983). I shall take a look at some of these ideas about "non-
von Neumann" style computing in the discussion of visual perception in
chapter 10.
Understanding of Language
In addition to this influx of new ways of thinking, there have also
been impressive achievements in some of the specific subject areas of
artificial intelligence-achievements that reflect new concepts and can be
said to atone for the contrived performances of the earlier generation of
programs. In the area of language, for example, Roger Schank and his
colleagues at Yale University, as well as his students who have now moved
to other research centers, have produced programs that give precis of
stories, skim newspaper articles, and answer questions and draw inferences
about plot, character, and motivation in stories (Schank and Abelson 1977;
Wilensky 1983).
Schank's claims about his programs, as well as his more general theo-
ries of language, have generated much controversy. He traces the failure
of early language programs to their narrow focus on grammar and calls for
programs that truly "understand the language," or at least parts of it. But
167
many critics feel that understanding is being used in an illegitimate sense. I

shall review this line of criticism when I come to John Searle's general
skepticism about machines engaging in any kind of "intentional" behavior.
There has also been debate about Schank's characterization of language.
He maintains that everything we can talk or think about boils down to a
small group of basic conceptual elements, and goes so far as to declare just
what those elements are and how they work, thus doing apparent violence
to subtleties of meaning. For instance, he claims that all the verbs of
everyday speech can be analyzed in terms of twelve primitive actions (for
example, move, ingest, grasp), which concern the handling, movement or
transference of things, abstract relationships, and ideas. Schank provides
no theoretical justification for his list but puts forth many examples of how
one can analyze verbs in light of these primitive cores. Upon these verbs
one should be able to construct a general understanding of language, one
focusing almost entirely on semantics, to the exclusion of syntactic factors.
Schank's linguistic formulations have not convinced those who work
in the Chomskian tradition of linguistic processing. From their perspective,
Schank has created a set of ad hoc mechanisms which, however successful
in certain limited circumstances, are completely unprincipled. There is no
systematic criterion for determining when a particular script should be
activated, or a theoretically motivated reason to choose one particular
conceptualization of a verb over another. For example, why analyze eating
in terms of ingesting, rather than including the muscular movements,
thoughts, and reactions of an agent who is eating food? Or why invoke
a script of a doctor's office rather than an office in general_ or an oral
surgeon's in particular, when going to a dentist? To such critics, there is
no point in proceeding with the task of language understanding until one
has a viable theory of the structure of language-a theory that includes the
various aspects of language, including syntax, and can be applied to any
kind of linguistic input (Dresher and Hornstein 1976).
Though many observers would agree on the theoretical limitations of
Schank's pursuit, the fact that his programs work reasonably well at their
appointed task cannot be readily dismissed. Moreover, many scholars be-
lieve that Schank has hit upon the level of generalization that may be
appropriate for the creation of an "understanding system." He focuses on
units of meanings, not individual words; uses words to retrieve expecta-
tions rather than to craft sentences; and focuses on semantics rather than
on syntax. Just as the top-down approach has helped to rejuvenate cogni-
tive psychology, Schank's strategic "bets" have proved surprisingly suc-
cessful, at least up to now.
While the prospect of a program that understands natural language as
human beings do still seems far off, there have been significant advances
168
in other aspects of natural-language competence. Some approaches, such
as those of William Woods (1975) and Ross Quillian (1968), follow the
road of semantics; while others, such as those of Mitchell Marcus (1979),
prefer the trail of syntax. Raj Reddy's HEARSAY (Reddy et al. 1973)
program addresses the problem of understanding speech by drawing on
several kinds of knowledge-semantic, pragmatic, syntactic, lexical, pho-
nemic, and phonetic-and analyzing the signal at a variety of levels from
the single-sound segment to the complete phrase. Each knowledge source
takes a look at the hypothesis generated by all the others (and "displayed"
on a central blackboard) and then makes a "best" guess about what has
been said. While this system certainly lacks elegance, or the tightness of
a theoretically motivated approach, it exploits that quality of piecing to-
gether bits of evidence that may be involved in understanding speech
under less than optimal circumstances.
Perception
There has been analogous, and perhaps more unambiguous, progress
in the area of visual perception. Around 1960 a brief flurry of excitement
surrounded PERCEPTRON, a mechanism designed by Cornell's Frank
Rosenblatt for the purpose of recognizing letters (and other patterns)
placed in front of its "eyes." PERCEPTRON consisted of a grid of four-
hundred photocells, corresponding to neurons in the retina; these cells
were connected to associator elements, whose function was to collect elec-
trical impulses produced by the photocells; connections were made ran-
domly, because it seemed the best way to mimic the brain (Bernstein 1981;
Dreyfus 1972). (When random wiring did not work, PERCEPTRON was
rewired in a more deliberate way, which improved its performance in
pattern recognition.) Another set of components entailed response units:
an associator would produce a signal only if the stimulus was above a
certain threshold.
Marvin Minsky and Seymour Papert at M.I.T. believed that the rna-
chine was built upon erroneous concepts, and that it was more profitable
to find the principles that will make a machine learn than to build one in
the hope that it just might work. As they saw it, it was necessary to build
some prior structure in the machine and to provide the system with in-
formative feedback about successes or failures. Minsky and Papert eventu-
ally published a book in which they demonstrated conclusively the limita-
tions of PERCEPTRON theory (1968). For a while their critique put a
damper on the field, because A.I. workers drew the conclusion that nothing
further could be accomplished in the area of form recognition using sys-
tems based on the model of neural networks.
169
I I I THE CoGNITivE SciENCES : HisTORICAL PERSPECTIVE
However, at M.I.T. by the early 1970s, more impressive efforts had

been launched in the area of visual perception. Patrick Winston's program
(which seems to have escaped an acronymous fate) was able to learn how
to distinguish those block configurations that were arches from those that
were not. The program learned to recognize tables, arches, pedestals, and
arcades by being shown examples and counterexamples of each. The pro-
gram searched for those differences between positive and negative in-
stances which affected proper identification. A tentatively posited concept
was then enriched by analyses of subsequent instances. While (like
SHRDLU) the Winston program worked only in a very narrow domain, it
made use of deep insights about how shapes can be juxtaposed.
Building on several other lines of work (Clowes 1971; Roberts 1965),
David Waltz (1975) at M.I.T. devised a program capable of analyzing an
entire graphic scene. Not only could it contrive a three-dimensional de-
scription of objects in a drawn scene, but it was also able to recognize two
different pictures as being representations of the same scene. Waltz ex-
ploited an empirical discovery: if you can label the elements of which an
image is constituted, you can so constrain the physically possible scenes
that, in most cases, there is but a single interpretation of the scene. In fact,
it was even possible to resolve ambiguity in the figure. As Patrick Winston
declared:
Waltz' work on understanding scenes surprised everyone. Previously it was

believed that only a program with a complicated control structure and lots of
explicit reasoning power could hope to analyze complicated scenes. Now we know
that understanding the constraints the real world imposes at junctions is enough
to make things much simpler. A table which contains a list of the few thousand
physically possible ways that edges come together and a simple matching program
are all that is required .... It is just a matter of executing a very simple constraint-
dependent iterative process that successively throws away incompatible line
arrangement combinations. {1977, p. 227)
Even more impressive accomplishments in the area of visual percep-

tion came from another M.I.T. researcher, David Marr, who took upon
himself the task of modeling the early phases of the perception of objects
and scenes. Marr was impressed by Chomsky's theoretical approach to
what a language is and how any organism can learn a language; accord-
ingly, Marr pondered what is involved in any artificial-intelligence pro-
gram and in any visual system as well as the actual details involved in
recognizing a complex scene in the real world. And while he did not
himself work directly with the visual nervous system, he designed his
programs to be consistent with the processes known to characterize the
visual system. Marr's sophistication in formulating the problem as well as
170
ArHficial Intelligence: The Expert Tool
certain striking results won him widespread admiration before his un-
timely death at the age of thirty-five in 1980. I shall take a closer look at
Marr's accomplishments (Marr 1982) when I focus in chapter 10 on the
issue of how one recognizes an object.
Of course, many other researchers and programmers could be men-
tioned: both those involved in "mainline" work in language, visual percep-
tion, and problem solving; and those tilling peripheral soils, such as Chris-
topher Longuet-Higgins's (1979) work in the recognition of musical keys
or Richard Young's (1974) work simulating children's cognitive processes
on Piaget-type tasks. There are also under way promising new efforts,
whose overall impact cannot yet be judged. For example, Douglas Lenat
(1976; 1984) has developed an approach to heuristic reasoning as a supple-
mentary means for solving problems and even for finding new heuristics.
In any event, the general point should be clear: there has been a second
burst of energy in the field of artificial intelligence since the crisis of the
early 1970s. If the accomplishments are not yet up to human snuff, they
can no longer be readily dismissed as resting on superficial procedures or
depending on cheap tricks, but are clearly addressing central issues in
human intelligent behavior.
The Chinese Room
Nonetheless, not all observers are convinced of the ultimate worth of

artificial intelligence. In what may well be the most searching critique of
it yet, the Berkeley philosopher of language John Searle has written an
article entitled "Minds, Brains, and Programs" (1980). Because Searle's
article was published in a journal of open peer commentary, where it
generated several dozen responses from sympathetic as well as angered
critics, the issues involved in the debate have become well known and, in
a sense, serve as a capsule summary of the principal themes in contempo-
rary artificial intelligence.
Searle begins by excluding from his critique the claim of weak or
cautious A.l.: that is, that artificial intelligence can illuminate the processes
of human behavior. He has no problem (at least as far as his article is
concerned) with this variant of A.I., where the computer serves as a tool
for the study of mind, allowing hypotheses to be rigorously tested. Searle's
quarrel lies with the claim of strong A.l., that the appropriately pro-
grammed computer really is a mind and can be said literally to understand
and to experience other cognitive states. While few scholars hold the
171
"strong" view as baldly as does Searle's hypothetical target, the vehemence

of certain responses to Searle's critique confirm that he struck a raw nerve.
Searles Conundrum
To combat the strong view of A.l., Searle introduces his provocative
Chinese room example. Suppose, he says, that an individual (in this case,
Searle) is locked in a room and given a large batch of Chinese writing.
Suppose that he knows no Chinese writing and may not even be able to
discriminate Chinese from other kinds of squiggles. Suppose next that he
is given a second set of Chinese characters together with a set of rules for
collating the second batch with the first. The rules are in English and are
therefore understood by Searle. The rules teach him to correlate one set of
formal symbols with another set. This process is repeated with other
materials so that the speaker becomes accomplished at correlating the
characters with one another and, hence, can always provide the "right" set
of characters when given an initiating set. Then, to complicate the situation
further, suppose that the speaker is also given questions and answers in
English, which he is also able to handle. The speaker is then faced with
the following situation:
Suppose also that after a while I get so good at following the instructions for
manipulating the Chinese symbols and the programmers get so good at writing the
programs that from the external point of view-that is, from the point of view of
somebody outside the room in which I am locked-my answers to the questions
are absolutely indistinguishable from those of native Chinese speakers. Nobody
just looking at my answers can tell that I don't speak a word of Chinese. Let us
also suppose that my answers to the English questions are, as they no doubt would
be, indistinguishable from those of other native English speakers, for the simple
reason that I am a native English speaker. From the external point of view-from
the point of view of someone reading my "answers"-the answers to the Chinese
questions and the English questions are equally good. But in the Chinese case,
unlike the English case, I produce the answers by manipulating uninterpreted
formal symbols. As far as the Chinese is concerned, I simply behave like a com-
puter: I perform computational operations on formally specific elements. For the
purposes of the Chinese, I am simply an instantiation of the computer program.
(Quoted in Hofstadter and Dennett 1981, p. 33)
Searle's argument is more complex and subtle than this single quota-
tion reveals, but its general tenor can be grasped. He believes that the
Turing test is no test whatsoever of whether a computer program has a
mind or is anything like a human being, because humanlike performance
can be faked by any individual or machine that has been supplied with a
set of formal rules to follow under specific circumstances. So long as one
172
is simply following rules-so long as one is engaged in syntactic operations
-one cannot be said to truly understand. And the computer is a machine
par excellence for formal operations, unencumbered by any semantic knowl-
edge, by any knowledge of the real world, or by any intention to achieve
certain effects by its specific response. Therefore, it is a fundamentally
different kind of entity from a human being who understands the semantic
content of an utterance, and has his own purposes for communicating.
Conclusion: the program of strong A.I. is bankrupt.
Counterattacks
Knowing that his argument will not go unchallenged, Searle himself
anticipates the major counterarguments. He reviews the Berkeley or "sys-
tems reply": the claim that while the individual in the room may not
understand the story, he is part of a whole system-ledger, paper, data
banks, and so on; and the system itself understands. Searle's response is
that the individual could memorize or internalize all the materials in the
system (for example, memorize the rules in the ledger and the data lists of
Chinese symbols and do all calculations in his head) and still would not
understand.
A second response is the robot response of the Yale Contingent.
According to this line of argument, a computer could be put inside a robot,
and the robot could walk around and do all the sorts of things that real
people do; being thus in touch with the real world, such a robot would
really understand. But Searle says that, while this reply tacitly agrees that
understanding is more than formal symbol manipulation, it still fails to
make its case-because Searle could, unbeknownst to himself, actually
become the robot. For example, the Chinese symbols might actually come
to him from a television camera attached to the robot: thus, the Searle-
robot would appear to be perceiving. Analogously, the symbols that he is
issuing forth could make motors move the robots, arms, and legs: thus, the
Searle-robot would act. In either case, however, Searle would actually be
making all of the formal manipulations of symbols by himself and still
would know nothing of what is going on. In one sense, he would be the
homunculus of the robot but would remain in total ignorance of the
Chinese meanings.
A third line of argument, traced this time to Berkeley and M.I.T., is
the brain-simulator reply. This response suggests another scenario: a pro-
gram now simulates the actual sequence of neuron firings at the synapses
of the brain in a native Chinese speaker when he is engaged in, for exam-
ple, story understanding or question answering. The machine now taking
in stories simulates the formal structure of actual Chinese brains in proc-
173
II I THE CoGNITivE SciENCES : HisTORICAL PERSPECTIVE
essing the stories, and gives out Chinese answers as output. Searle says that
even creeping up this close to the operation of the brain still does not
suffice to produce understanding. Instead, he says, imagine that rather than
a monolingual man being in a room shuffling symbols, the man is equipped
with an elaborate set of water pipes with valves connected to them. When
the man receives the symbols, he looks them up in the program and then
determines which valves to turn on and off. Each water connection corre-
sponds to a synapse in the Chinese brain, and the whole system is so rigged
that eventually Chinese answers pop out. Now, points out Searle, the man
does not understand Chinese, nor do the water pipes. The problem with
the brain simulator is that it is simulating the wrong things about the brain.
Searle indicates that
as long as it simulates only the formal structure of the sequence of neuron

firings at the synapses, it won't have simulated what matters about the brain,
namely its causal properties, its ability to produce intentional states.... That the
formal properties are not sufficient for all the causal properties is shown by the
water pipe example: we can have all the formal properties carved off from the
relevant neurobiological causal properties. (Quoted in Hofstadter and Dennett
1981, p. 340)
Searle goes on to list other replies, and combinations of replies, and

responds to them in similar fashion, but it is in his response to the brain
simulator reply that he tips his hand. He believes understanding is a
property that comes from a certain kind of machine only-a machine like
the human brain which is capable of certain processes, such as having and
realizing intentions. As he says, it is not because I am an instantiation of
a computer program that I am able to understand English; but, as far as we
know, "it is because I am a certain sort of organism with a certain biological
(i.e., chemical and physical) structure, and this structure, under certain
conditions, is causally capable of producing perception, action, under-
standing, learning, and other intentional phenomena" (quoted in Hof-
stadter and Dennett 1981, p. 344).
In the journal, respondents attacked Searle's argument at every point.
Some, steadfastly embracing positions Searle has already attacked, said
that he had misunderstood them, or that when properly pursued, particu-
lar responses could indeed accomplish what Searle denied. Other critics
accused him of being a mystic, antiscientific, antitechnological, or just
another philosopher gone bad. Much was made about the caricature em-
bodied in his single fake Chinese speaker. Respondents claimed that
Searle's example is just not plausible: in order to be able to carry off the
behavior successfully, the man or the system would need to achieve genu-
ine understanding, or what we would unhesitatingly call genuine under-
174
Artificial Intelligmce: The Expert Tool
standing were it to be found in a human being. There are orders of magni-
tude Searle ignored (for example, the effort it would take to anticipate and
translate all possible messages) and only this sleight-of-hand allows him
to make his case. Searle has also been accused of confusing levels: it is the
person who understands, and not the brain; and similarly, it is the com-
puter as a whole that understands, and not the program.
To my mind, when Searle remains on his own ground-in putting
forth the example of the man equipped with a set of formal rules for
responding in Chinese-he is convincing. While I would not join Searle in
ruling out the possibility of any machine's understanding a natural lan-
guage, I certainly feel that the word understand has been unduly stretched
in the case of the Chinese room problem in its various manifestations.
It is Searle's positive explanation of why the rest of us understand,
when he speaks about the intentionality of the brain, that can be legiti-
mately attacked. The whole notion of the brain as a causal system that
displays intentionality is obscure and difficult to understand, let alone to
lay out coolly like a computer program. Zenon Pylyshyn, a computer
scientist from the University of Western Ontario, scores telling points:
"What Searle wants to claim is that only systems that are equivalent to
humans ... can have intentionality. His thesis thus hangs on the assump-
tion that intentionality is tied very closely to specific material properties
-indeed, that it is literally caused by them." Pylyshyn wonders whether
Searle is proposing that intentionality is a substance secreted by the brain
(1980a ), and then poses this puzzle:
[Suppose] if more and more of the cells in your brain were to be replaced by
integrated circuit chips, programmed in such a way as to keep the input-out
function of each unit identical to that of the unit being replaced, you would in all
likelihood just keep right on speaking exactly as you are doing now except that
you would eventually stop meaning anything by it. What we outside observers
might take to be words would become for you just certain noises that circuits
caused you to make. (P. 442)
Searle's argument, it seems to me,.loses its force if, by definition, only

human brain or brainlike mechanisms can exhibit properties of intention-
ality, understanding, and the like. If this is true by definition, then there
is no point to the controversy. If, on the other hand, Searle allows (as he
must) that nonprotoplasmic entities can also possess the "milk of human
intentionality," he must explain what it takes to be intentional, to possess
understanding, and the like. Such explanation is going to be difficult to
provide, because we have no idea how the causal properties of protoplasm
allow individuals to think; and, for all we know, the process is as odd as
one of those Searle so effectively ridicules.
175
In an effort to address these concerns, Searle has written a book on

Intentionality which he defines as the property of mental states and events
by which they are directed at objects and states of affairs in the world.
Intentional states include beliefs, fears, desires, and intentions (1983a,
p. 1). In the book, he declares:
My own speculation, and at the present state of our knowledge of neuro-

physiology it can only be a speculation, is that if we come to understand the
operation of the brain in producing Intentionality, it is likely to be on principles
that are quite different from those we now employ, as different as the principles
of quantum mechanics are from the principles of Newtonian mechanics; but any
principles, to give us an adequate account of the brain, will have to recognize the
reality of, and explain the causal capacities of, the Intentionality of the brain.
(P. 272}
Indeed, Searle seems to believe that, to do cognitive studies, one needs only
two levels of explanation-the level of intentionality (a plain English
discussion of the organism's wishes, beliefs, and so on)-and a neuro-
physiological explanation of what the brain does in realizing these inten-
tional states. He finds no need for a level of symbolic representation, which
dominates work throughout the cognitive sciences (1983b). To the extent
that these lines of argument sound like mere handwaving, Searle's positive
assertions about intentionality have little force. To the extent that he can
specify just what he means, and show that intentionality is not in principle
restricted to organic brains, a genuine scientific issue will have been joined.
Thereafter, researchers in artificial intelligence can try to simulate the very
behavior that Searle believes lies beyond their explanatory scope.
It is possible, however, that the issues between Searle and his critics
-and perhaps even between those in sympathy and those out of sympa-
thy with A.I.-transcend questions of a scientific nature. In an angry
critique of the Searle paper, computer scientist Douglas Hofstadter makes
the following conjecture:
This religious diatribe against AI, masquerading as a serious scientific argu-

ment, is one of the wrongest, most infuriating articles I have ever read in my life.
. . . I know that this journal is not the place for philosophical and religious
commentary, yet it seems to me that what Searle and I have is, at the deepest level,
a religious disagreement and I doubt that anything I say could ever change his
mind. He insists on things he calls "causal intentional properties" which seem to
vanish as soon as you analyze them, find rules for them, or simulate them. But what
those things are, other than epiphenomena, or innocently emergent qualities I don't
know. (Quoted in Searle 1980, p. 434)
If Hofstadter is right about the underlying reasons for the disagreement,

then there is no way in which either side can convince the other. At stake
176
is a matter of faith, in which reason will not even be considered relevant
by either side.
Critics and Defenders: The Debate Continues
Whatever the merit of Searle's arguments, they do not question the legiti-
macy of most current work in artificial intelligence; in fact, as he indicates,
he has no qualms about accepting the utility of A.l. as an aid in conceptual-
izing and testing theories of human intelligence. But, as we have seen,
some critics challenge A.I. at an even more fundamental level. As one
example, Joseph Weizenbaum (1976), raises ethical questions about
whether A.l. should be allowed to impinge on territory that has hitherto
been restricted to human beings, such as issues of justice or love.
Another line of criticism suggests that efforts to simulate human intel-
ligence are perfectly valid, but that most of the A .I. community has hereto-
fore used superficial models which do not approach the core of human
thought processes. Initially, this line of criticism came from those in the
camp of transformational linguistics, like Elan Dresher, Norbert Hornstein
(Dresher and Hornstein 1976), and Kenneth Wexler (1978). From their
perspective, rather than trying to build upon deep principles of language
(or whatever area happens to be modeled), artificial intelligence is simply
a practical, trial-and-error pursuit and is unlikely ever to come up with
anything that has generality and explanatory power. For instance, as they
see it, Schank's program is too vague and unspecified, and there are no
algorithms for determining the "primitive verb" or the script that is rele-
vant to a given linguistic string. In contrast, Winograd's program is criti-
cized for being too specifically tied to a given "micro-world" and thus
being inapplicable to any new specimen of language. According to this
"purist" philosophy, scientists ought to be spending their time analyzing
systems in themselves, studying competent human beings, or fashioning
theoretical explanations of what intelligence or competence amounts to:
theories that can stand up to the standards of the "hard" sciences. Perhaps,
once the real operative principles are understood-as they are beginning to
be understood in the area of syntactic processing-it should be relatively
trivial to program the computer to carry out these operations. In the next
chapter, I shall examine this transformationalist vision of acceptable ex-
planatory models in cognitive science.
These critics are arguing that artificial intelligence has not yet attained
the standing of a genuine scientific endeavor. Instead, it is just a convenient
177
tool whose operating principles are of scant theoretical interest-views

that have been recently echoed by some of the most active workers in
artificial intelligence. Terry Winograd, once the darling of the A.I. commu-
nity, has been pondering the limitations of computer-based understanding
and has moved away from simulation experiments. He and Daniel Bobrow
declare:
Current systems, even the best ones, often resemble a house of cards .... The
result is an extremely fragile structure which may reach impressive heights, but
collapses immediately if swayed in the slightest from the specific domain (often
even the specific examples) for which it was built. (Bobrow and Winograd 1977,
p. 4)
Schank has conceded the difficulties involved in devising systems without

having good models, and has newly acknowledged the philosophical
nature of many problems:
AI is very hard. What is the nature of knowledge? How do you abstract from
existing knowledge to more general knowledge? How do you modify knowledge
when you fail? Are there principles of problem solving that are independent of
domain? How do goals and plans relate to understanding? The computer is a way
of testing out ideas. But first, we need to understand what we're supposed to be
building models of. (Quoted in Waldrop 1984a, p. 805)
Workers like John McCarthy (1984) have been pondering what Daniel
Dennett (1983) terms the "smoking gun" problem of artificial intelligence:
how a program can decide which of the unexpected features in a situation
can be ignored. (For example, if it is to resemble human intelligence, a
frame of a boat trip should ignore the fact that the oars on a rowboat are
blue rather than green but dare not ignore the fact that one of the oarlocks
is broken.) John Seely Brown and Kurt van Lehn, researchers at Xerox
P ARC in Palo Alto, admit that most current A. I. work fails to meet tradi-
tional criteria of scientific theories, and decry the absence of "competitive
argumentation" whereby the power of one simulation can be rigorously
compared with the power of another. By such comparison, it is possible
to indicate why one accepts some computational principles, while rejecting
others. In lieu of such argumentation, these authors wryly suggest, artifi-
cial intelligence theories have stood on the toes rather than on the shoul-
ders of their predecessors (van Lehn, Seely Brown, and Greeno 1982).
The realization that one's chosen discipline may not have been operat-
ing in a scientifically impeccable manner is not in itself an instant panacea,
but it is an important first step-maybe the most important step in the
maturing of a discipline. A second valuable trend, in my view, has been
the growing number of workers in A.I. who seek systems reflecting the
178
deep principles at work in a particular domain of knowledge. Spurning
both the search for the most general properties of problem solving (which
proved elusive with the General Problem Solver) and the interest in those
expert systems that perform well simply by drawing with brute force on
massive amounts of stored knowledge, these researchers take seriously the
principles that appear to be at work in the only other intelligent system
we know-the human being. I shall examine some of their promising
efforts in the latter chapters of this book.
It is possible, of course, as a result of a careful study of different areas,
we will discover that the deepest principles do indeed operate across areas,
and that we will ultimately require a general knowledge base or problem
solver rather than a host of separate experts. As I suggested with respect
to the work of John Anderson (1983) in cognitive psychology, such a
dream may still materialize, but the likelihood seems small.
My own analysis suggests that, after a period of excessive claims and
sometimes superficial demonstrations, artificial intelligence has advanced
to a more measured view of itself and has, in the process, attained a set of
reasonably solid accomplishments. This maturing process has involved a
recognition that the practice of A.I. entails deep philosophical issues which
cannot be ignored or minimized. Involved as well has been a recognition
that there are limits to what can be explained by current A.I. methods and
that even whole areas of study may lie outside of artificial intelligence, at
least now and perhaps permanently. It is important to have genuine demon-
strations and not just verbal descriptions of possible programs: this insis-
tence has been among the greatest contributions of Newell and Simon. But
it is equally important that these demonstrations reflect robust principles
and are not just fragile constructions with limited application.
Given my own view that the future of cognitive science lies in the
study and elucidation of particular domains of knowledge, I believe that
the field will achieve scientific status when it freshly illuminates the do-
mains and knowledge systems with which it is concerned. As the psy-
chologist John Marshall declares:
I just don't care what members of the AI community think about the ontologi-
cal status of their creations. What I do care about is whether anyone can produce
principled, revealing accounts of, say, the perception of tonal music ... , the
properties of stereo vision ... , and the parsing of natural language sentences.
Everyone that I know who tinkers around with computers does so because he has
an attractive theory of some psychological capacity and wishes to explore certain
consequences of the theory algorithmically. (1980, p. 436)
Christopher Longuet-Higgins makes the same general point in a

somewhat different way:
179
It is perhaps time that the title "artificial intelligence" were replaced by

something more modest and less provisional. . . . Might one suggest, with due
deference to the psychological community, that "theoretical psychology" is really
the right heading under which to classify artificial intelligence studies of perception
and cognition .... The task of the theoretician is to formulate hypotheses and to
elicit their logical implications as carefully as he can, with due attention to matters
of internal consistency and predictive power; the duty of the experimenter is to
confront the predictions of a theory with firm and relevant observations, and to
suggest points at which the theory needs modifying in order to bring it into line
with experiment-if that is indeed possible. The time has now come, it seems,
when the task of theory construction is altogether too intricate to be consigned to
spare moments away from the laboratory; it is at least as much of a discipline as
good experimentation. (1981, p. 200)
Marshall and Longuet-Higgins are pointing to the increasingly close

ties being forged between experimental cognitive psychology and artificial
intelligence. Psychologists can benefit from the careful simulations made
by A.I. researchers, and can put their own typically informal models to
rigorous tests; A.I. scientists can determine whether their hypothesized
models of human behavior are actually realized by the subjects about
whom they have been speculating. It seems plausible to me that parts of
psychology and parts of computer science will simply merge into a single
discipline or, as I suggested in the previous chapter, that they will form the
central core of a newly forged cognitive science.
Of course, there may well be areas of computer science, as well as of
psychology, that do not take part in this merger. This would be quite
proper. The idea that artificial intelligence must be able to handle all
psychological (or all philosophical) issues, or it will be unable to handle
any of them, must be exorcised. Similarly, there is no more reason for us
to think of humans as being completely identical to computers than for us
to cling to the notion that there are no useful resemblances or parallels
between these two kinds of (potentially) thoughtful entity.
Nonetheless, the issue of the actual degree of similarity between hu-
mans and computers cannot be permanently ignored. One of the principal
findings of recent artificial intelligence is that the digital serial von Neu-
mann computer seems in many ways an inadequate model of human
cognition. To the extent that this finding is reinforced-and much of the
research in the latter part of this book can be cited as supporting evidence
-then we will have confirmation of the computational paradox. Artificial
intelligence has demonstrated that the computer can be a useful tool for
studying cognition and that it serves as a reasonable model for some
human thought processes. But whether it is the best model for the most
important processes is still very much an open question.
Artificial intelligence and cognitive psychology may well merge as the
180
central components of a new cognitive science, but they cannot in them-
selves constitute the field. Philosophy will remain a source of important
questions, a repository of invaluable reflections on those questions, and a
valuable critic of cognitive scientific practices. But an equally important
partner in this newly fashioned line of inquiry will be those disciplines that
take as their appointed task the analysis of particular domains of cognition.
Of the various disciplines that must assume this burden, linguistics is, by
all accounts, the discipline that has accomplished the most and is thus most
centrally involved in cognitive science. Moreover, and perhaps for this
reason, other disciplines look to linguistics as a possible model of how to
study a particular domain and to apply the powerful (and perhaps too
powerful) tools of the psychologist and the computer scientist. Because of
the intrinsic importance of language among the human cognitive faculties,
because of linguistics' role as a possible model for domain-specific studies,
and because of the tremendous strides made in linguistics over the past few
decades, it is appropriate to turn now to the scientific study of language.
181
7
Linguistics: The Search

for Autonomy
At First, There Were Colorless Green Ideas ...
Enigmatic Sentences
At first reading, one of the most famous sentences of twentieth-
century science makes little everyday sense. This is Noam Chomsky's
oxymoronic proposition "Colorless green ideas sleep furiously." After all,
ideas don't have colors, ideas don't sleep, something can't be simultane-
ously colorless and green-and for that matter, what does it mean for any
entity to sleep furiously?
And yet a certain sense can be squeezed out of this sentence. It is
certainly clearer than "ideas furiously green colorless sleep" or some other
random concatenation of these five substantives. In Chomsky's sentence,
one knows that a proposition is being asserted about ideas; one senses a
contrast being drawn between an apparent state of blandness, as against
a more active state of energy and color. The sentence is even susceptible
to a figurative reading. One thinks of the revolutionary impact of appar-
ently innocuous ideas-the kinds of idea Noam Chomsky himself put
forth in his 1957 monograph Syntactic Structures.
The fact that we have clear intuitions about apparent nonsense like
"Colorless green ideas sleep furiously" (or "Twas brillig and the slithy
toves did gyre and gimble") underlies Chomsky's central contributions to
linguistics and, for that matter, to cognitive science in a broader sense.
182
Linguistics: The Search for Autonomy
What Chomsky accomplished in his monograph, and in the numerous
writings that followed, was to call attention to certain properties of sent-
ences which all normal speakers and hearers intuitively know, but which
derive from a deeper understanding of the language whose properties may
be explicitly known only to the linguist. Chomsky called attention to and
offered a convincing explanation of differences between apparently similar
sentences: "John is easy to please" (where John is the recipient of the
pleasing) versus "John is eager to please" (where John does the pleasing).
"Ted persuaded John to learn" (where John is to do the learning) versus
"Ted promised John to learn" (where Ted does the learning). Chomsky
provided an account of sentences where meaning can be preserved despite
shifting of the principal terms ("The cat chased the mouse" versus "The
mouse was chased by the cat") as against those sentence pairs where one
cannot simply reverse clauses ("Many men read few books" versus "Few
books are read by many men"). He singled out and suggested mechanisms
underlying the human ability to detect and unravel the ambiguities con-
tained in such sentences as "Flying planes can be dangerous," "The shoot-
ing of the hunters disturbed me," and "I didn't shoot John because I like
him." And he marked for study sentences that seem acceptable: "John
seemed to each of the men to like the other" as against those superficially
similar sentences that violate some putative rule of the language: "John
seems to the men to like each other." We can say "We like each other"
meaning each of us likes the other, but we can't say "We expect John to
like each other"-indeed we're not even sure what this could mean
(Marshall 1981).
That human beings have intuitions for sentences like these was not
an insight unique to Chomsky. After all, Lewis Carroll exploited similar
knowledge in fabberwoclcy. The influential linguist of the early part of this
century, Edward Sapir (1921), had called attention to many of the very
relationships among sente~ces that were later to occupy Chomsky. But
Chomsky went beyond his predecessors, including insightful linguists like
Sapir, in his goal of setting out the rules that allow individuals to make,
or to generate, all of the correct sentences listed above; to know that they
are correct and what they mean; and to be able to pick out those sentences
that violate these rules and are hence ungrammatical, though not necessar-
ily devoid of meaning. To do this, the speaker must possess at some
level a detailed set of rules or procedures indicating when different parts of
speech can occur in given places in an utterance: that is, the rules must
capture the intuitions of native speakers about the relations obtaining
within and among sentences. Chomsky's statement of this explicit goal for
the study of language-or, more precisely, for the study of syntax-and
his success in developing methods directed toward the achievement of this
183
I I I THE CoGNITIVE SciENCES : HisToRICAL PERSPECTIVE
goal instantly made his work stand out from that of other students of
language. And his more general conviction that the several domains of
mind (such as language) operate in terms of rules or principles that can be
ferreted out and stated formally constitutes his main challenge to contem-
porary cognitive science.
In Syntactic Structures, Chomsky first announced a set of goals and then
described the kind of grammar that appeared necessary if the proper
regularities in English-or, indeed, in any language-were to be discerned.
Rather than simply looking at the data of language, and trying to discern
regularities from empirically observed utterances, as his predecessors had
typically done, Chomsky insisted that the principles would never emerge
from a study of the utterances themselves. Instead, it was necessary to
work deductively. One must figure out what kind of a system language is,
much in the way that one figures out what a particular branch of mathe-
matics is like; and one must state one's conclusions in terms of a formal
system. Such an analysis should lead to positing of the rules that can
account for the production of any conceivable grammatical sentence (and
there is, of course, an infinite number of such sentences), but at the same
time the rules should not "generate" any incorrect or ungrammatical sen-
tences. Once the system has been set up, one should then examine particu-
lar utterances to determine whether they can, in fact, be appropriately
generated through adherence to the rules of the linguistic system.
In embarking on this program, Chomsky made two important, simpli-
fying assumptions. One assumption was that the syntax of language could
be examined independently of other aspects of language. If syntax were
inextricably bound up with other aspects of language-for example, with
meaning or with communicative utility-then it might not be possible to
figure out its governing laws. The second but closely related tacit assump-
tion was that the discipline of linguistics could proceed independently of
other areas of the cognitive sciences. Once again, if the study of language
were integrally tied to the study of other areas of human cognition-to
other mental organs, as Chomsky would later phrase it-then progress
might be impossible or agonizingly slow.
For the most part, these working assumptions about linguistic auton-
omy-the autonomy both of syntax from other aspects of language, and
of linguistics from other aspects of cognitive science-worked out propi-
tiously and made linguistics a rapidly developing area of science. But
whether these assumptions can ultimately be sustained-whether the as-
sumptions of autonomy are truly justified-constitutes a problem that has
yet to be resolved. I shall return to these issues in the concluding section,
after reviewing Chomsky's principal discoveries, the history of the field
prior to his own entry, and certain current issues in the science of language.
184
Before entering into the heart of Chomsky's contributions, it may be
well to say a word about the focus of this chapter. While the work of other
scholars has been central in other chapters, in no other chapter of this book
has so much attention been focused on a single individual. In part, this is
an expositional device-a way of presenting the principal (and often com-
plicated) ideas of modern linguistics in as accessible a fashion as possible.
But, also, in no other contemporary cognitive science is the work of a single
individual so key and so irreplaceable. In a nontrivial sense, the history of
modern linguistics is the history of Chomsky's ideas and of the diverse
reactions to them on the part of the community.
s
Chomsky Approach
To achieve his ambitious goal, Chomsky first had to show that current
methods for analyzing syntax and accounting for acceptable sentences
would not work. By pushing a precise but inadequate formulation to its
unacceptable conclusion, one could show why that formulation was inade-
quate-and, in the process, obtain a deeper understanding of the linguistic
data it had vainly sought to explain. Chomsky proceeded by proving that
the theoretically most plausible method for generating sentences could not
in principle work. To begin with, he considered a finite-state grammar: a
machine with a finite number of internal states which generates sentences
in the following manner. Starting in a unique initial state, the machine
passes into a second state by producing the first word of a sentence; then,
while emitting a word at each transition, the machine proceeds from state
to state until it has reached the final state, at which time a complete
sentence has been generated. Chomsky showed that such finite-state
grammars are inherently incapable of representing the recursive properties
of English constructions: that is, by its nature, a finite-state grammar
cannot generate sentences in which one clause is embedded in or depen-
dent upon another, while simultaneously excluding strings that contradict
these dependencies.
As an example, consider the sentence "The man who said he would
help us is arriving today." Finite-state grammars cannot capture the struc-
tural link between man and is arriving which spans the intervening relative
clause. Moreover, as its name suggests, a finite-state grammar cannot han-
dle linguistic structures that can recur indefinitely, such as the embedding
of a clause within another clause ("the boy that the girl that the dog
... ," and so on). Even though such sentences soon become unwieldy for
the perceiver, they are strictly speaking grammatical: any grammar must
be able to account for (or generate) them. At the most general level, English
(and, for that matter, other languages) does not work by slotting a word
185
and then indicating what words can follow "to the right" of it; it works
at a higher level of abstraction, where certain elements, under certain
circumstances, can be placed wholly within other elements. Thus,
Chomsky was making (with respect to language) precisely the point that
Karl Lashley had urged about serial behavior in all of its manifestations.
Chomsky also demonstrated the unwieldiness-though not the utter
impossibility--of a second kind of grammar, a phrase-structure grammar (or an
immediate constituent analysis). Here the linguist works with an initial set
of strings, together with a finite set of phrase-structure or "rewrite rules,"
where one phrase can be rewritten in another permissible form. Chomsky
showed that such a phrase-structure system can only generate with great
unwieldiness certain sentences; moreover, it cannot capture or account for
many of the regularities any English speaker appreciates. Thus a phrase-
structure grammar cannot explain the different structures of "What are
you looking for" and "What are you running for" (Lees 1957, p. 386), or
the ambiguity of "The teacher's marks are very low." Nor is there any way
of showing that "John hit Bill" and "Bill is hit by John" bear a close
relationship to one another. Nor, according to Chomsky, is it possible,
using a phrase structure grammar, to generate sentences that involve the
combination of two parallel clauses: to be specific, such grammars offer no
mechanism for combining sentences like "The scene of the movie was in
Chicago" and "The scene of the play was in Chicago" into "The scene of
the movie and of the play was in Chicago." In broader terms, the phrase-
structure grammar can be written only at the cost of several restatements
of extensive parts of the grammar. As linguist Robert Lees points out, "if
these uneconomical repetitions are permitted, then the grammar fails to
state the near identity in structure between those parts which must be
repeated" (1957, p. 387). More generally, while English might conceivably
be described in terms of phrase-structure grammars, such a description
would be so unwieldy, so hopelessly complex that these descriptions
would be of scant interest.
Thus, the most plausible models for the generation of grammatical
sentences were both shown to be inadequate: finite-state grammars, be-
cause countless instances of meaningful language cannot be generated on
a word-by-word basis; phrase-structure grammars, because mere attention
to the ways that phrases are constructed fails to capture important
regularities in the language. For these reasons, Chomsky found it necessary
to introduce a new level of linguistic structures which at once swept away
these difficulties and made it possible to account for the full range of
sentences in the language.
Inspired in part by his teacher Zellig Harris (1952), Chomsky discov-
ered the level of transformational analysis. In a transformational grammar,
186
Chomsky argued, one posits a set of rules whereby sentences can be related
to one another, and where one sentence (or more precisely, the abstract
representation of a sentence) can be converted or transformed into another.
A generative grammar, in Chomsky's sense, is a rule system formalized
with mathematical precision: without drawing upon any information that
is not represented explicitly within it, the system generates the grammati-
cal sentences of the language that it describes or characterizes, and assigns
to each sentence a structural description or grammatical analysis.
As described in Chomsky's 1957 monograph, the method of transfor-
mational grammar works roughly as follows. (It must be immediately
added-indeed, stressed-that the system has changed several times in the
intervening years.) Beginning with phrase-structure rules, one generates
only the cores of sentences, or kernel sentences, which are short active declar-
ative assertions. These are generated by following a set of instructions for
constructing strings: for example, (1) Sentence = Noun Phrase + Verb
Phrase; (2) Noun Phrase = T + N; (3) T = the; (4) N = man, ball; (5)
Verb Phrase = Verb + Noun Phrase; (6) Verb = hit, take; and so on.
Starting with the single symbol S, one can generate, by a completely
specifiable set of rules, a kernel sentence like "The man hit the ball."
Thereafter, all the other grammatical sentences of the language can be
generated by means of transforming these kernel sentences.
Transformations are an algorithmic set of procedures that occur in a
prescribed order and allow one to convert one linguistic string to another.
Thus, a transformation allows one to convert an active sentence into a
passive sentence, a positive statement into a negative statement or a ques-
tion. On such an analysis, "What are you looking for" can be described
as the "what-question" transformation of "You are looking for it"; while
"What are you running for" is a "why-question" transformation of "You
are running." Transformations can lay bare the links between sentences
like "The boy kissed the girl" and "The girl was kissed by the boy"; the
deep differences between superficially similar syntactic arrangements like
"The girl is eager to please" and "The girl is easy to please"; and the fact
that a phrase like "the shooting of the hunters" is ambiguous because
it can be accounted for by two different transformational histories-
one, in which the hunters are shot; the other, in which the hunters do the
shooting.
All these transformations are structure-dependent: that is, they do not
operate on single words or on strings of words of arbitrary length. Rather,
the transformations are imposed on strings (abstract representations of
sentences) after these strings have been analyzed into appropriate syntactic
categories and constituents (or phrase structures), which determine when
and where the transformations can be applied. Because Chomsky placed
187
great value on formal simplicity in the selection of the appropriate syntac-

tic description, transformational grammar won out over phrase structure.
While all English sentences can, in principle, be generated by a grammar
with only phrase-structure rules, one would naturally prefer a simple
phrase-structure rule and a simple transformational rule over a large num-
ber of cumbersome and difficult-to-orchestrate phrase-structure rules.
(To prevent subsequent misunderstanding, I must now get a bit ahead
of my story. At the time Chomsky wrote, it was not appreciated that the
transformation was an extremely powerful formal device, one that could
be abused to rescue an analysis which was not actually appropriate to the
linguistic data. To counter these and other problems, there has since been
a movement to reduce radically or even to eliminate the transformational
component. Also, several prominent linguists are once again re-embracing
a phrase-structure approach.)
While the transformational approach was clearly innovative, it is not
easy to indicate precisely where Chomsky's work went beyond that of his
contemporaries. Certainly the desire to write a complete grammar for
English had motivated others of his colleagues, and the focus on syntax,
to the relative exclusion of phonology and semantics, also characterized
much work in that era. Chomsky did, in fact, talk about how the syntactic
level interacted with phonology and semantics; but for years he has con-
centrated his analytic energies on the level of syntax (see, however,
Chomsky and Halle 1968). Moreover, Chomsky approached this task with
a seriousness of purpose, an arsenal of logical and mathematical tools, a
finesse and a finality of argument that had simply not been marshaled
hitherto in linguistic analysis. While other scholars had alluded to limita-
tions of existing models, like the finite-state or the phrase-structure model,
Chomsky actually proved the impossibility of the first and the essential
inadequacy of the second. While others, including Chomsky's teacher
Zellig Harris, had spoken about transformational rules that might link
sentences, Chomsky went beyond a mere assertion that such rules might
regularize relations among sentences: he asserted that there must be a
separate level called the transformational level, and he postulated the
mechanisms governing both obligatory and optional transformations.
Chomsky saw the linguist's task differently from the way his pre-
decessors had seen it. His view of grammatical generation was based on
the notion of an automaton-a machine in an abstract sense which simply
generates linguistic strings on the basis of rules that have been built (pro-
grammed) into it. The resulting grammar is neutral-equally valid as a
description of linguistic production or linguistic comprehension. Clearly,
Chomsky was a child of the new era of Wiener, von Neumann, Turing, and
Shannon, though-just as clearly-some of his specific ideas about how
language works ran directly counter to information-theory notions.
188
Chomsky also characterized the task of the linguist in a more explicit
way than had his predecessors. He devoted considerable attention to the
theoretical question of how one goes about choosing one linguistic model
over another. He laid out the formal criteria for an adequate theory of
linguistics suggesting (and demonstrating) how these criteria might be
achieved. He expounded as well an ordered set of standards of adequacy:
observational adequacy, where the grammar gives a correct account of observed
linguistic data; descriptive adequacy, where the account also captures the na-
tive speaker's intrinsic competence, his own internalized knowledge; and
explanatory adequacy, where the analyst uncovers principles underlying opti-
mal grammar for the language. An account that exhibits explanatory ade-
quacy can explain why individuals have constructed the grammars that
they have, and what kinds of abilities individuals have needed to achieve
linguistic competence. Such theoretical sophistication (and presumption!)
had also been hitherto absent from the writings of linguists interested in
grammar.
The Spreading of Green Ideas

Left entirely on its own, Chomsky's book might have proved too
remote from the rest of the field and perished on the vine. No one was
interested in publishing his massive doctoral thesis, The Logical Structure of
Linguistic Theory (1955), in which the formal proofs were laid out in detail.
His initial monograph, Syntactic Structures, was published by the small Dutch
publisher, Mouton, only after it had been strongly recommended by
Roman Jakobson, the most distinguished linguist of the day. Even so
issued, the monograph might well have remained obscure, at least for some
years, were it not for a long, detailed, and wholly positive review in the
influential journal Language by Chomsky's early student and associate
Robert Lees (1957).
Leaving no doubt in his review that Chomsky's slim volume would
revolutionize linguistics, Lees contrasted the "prescientific stage of col-
lection and classification of interesting facts" with a scientific discipline
"characterized essentially by the introduction of abstract constructs in
theories and the validation of those theories by testing their predictive
power" (1957, p. 376). Playing Huxley to Chomsky's Darwin, Lees as-
serted that "some fields of knowledge have reached such an advanced
stage of development that their basic problem can be stated very suc-
cinctly, and the structure is so well understood that we can now begin to
construct axiomatic theories to render explicitly and rigorously just what
their content is and means" (p. 377). In Lees's view, Chomsky's book
enabled linguistics to cross the line from a descriptive pre-science to an
axiomatic science:
189
Chomsky's book on syntactic structures is one of the first serious attempts on

the part of a linguist to construct within the tradition of scientific theory-construc-
tion a comprehensive theory of language which may be understood in the same
sense that a chemical, biological theory is ordinarily understood by experts in those
fields. It is not a mere reorganization of the data into a new kind of library
catalogue, nor another speculative philosophy about the nature of Man and Lan-
guage, but rather a rigorous explication of our intuitions about our language in
terms of an overt axiom system, the theorems derivable from it, explicit results
which may be compared with new data and other intuitions, all based plainly on
an overt theory of the internal structure of languages; and it may well provide an
opportunity for the application of explicit measures of simplicity to decide
preference of one form over another form of grammar. (1957, pp. 377-78)
Lees said that Syntactic Structures was one of the first serious attempts
to construct a comprehensive theory of language; but, as Frederic New-
meyer declares in his useful history of this era, "Actually the tone of the
review as a whole made it clear that Lees regarded it as the ONLY serious
attempt and a completely successful one at that" (1980, p. 19).
As they became aware of Chomsky's work, other linguists also dis-
cerned its power and originality. In another review, C. F. Voegelin, a
leading linguist, declared that even if Syntactic Structures managed to achieve
only part of its goals, "it will have accomplished a Copernican revolution"
(quoted in Newmeyer 1980, p. 19). The distinguished British linguist C. E.
Balzell remarked that "linguistics will never be the same"; and Bernard
Bloch, one of the leading linguists of an earlier generation, confided to
colleagues "Chomsky really seems to be on the right track. If I were
younger, I'd be on his bandwagon too" (both quoted in Newmeyer 1980,
p. 47).
But in heralding a revolution, Chomsky was launching a movement
at someone's expense-namely, the structural linguists of an earlier gener-
ation. And, as the full dimensions of his program became clear, there was
considerable and often violent opposition. For it turned out that Chomsky
was not merely offering a critique of existing attempts to describe a gram-
mar; instead, he espoused radically different views of how linguistics
should work, and, indeed, of what social science (we would today substi-
tute "cognitive science") was about. Some of these views were muted, or
had not yet been well formed, at the time of the publication of Syntactic
Structures (Chomsky was only twenty-nine at the time of the book's publi-
cation; twenty-seven at the conclusion of his doctorate); but with the
publication of succeeding articles in the late 1950s and early 1960s, and the
issuing of a major theoretical statement, Aspects of a Theory of Syntax, in 1965,
the full dimensions of his program had been laid out.
To begin with, Chomsky made it clear that he was interested in
language in a particular sense. Language is not a general means of commu-
190
nication or a blanket word encompassing all symbol systems: rather, at the
core of language is the property of syntax, the capacity unique to the
human species to combine and recombine verbal symbols in certain spe-
cifiable orders, in order to create a potentially infinite number of grammati-
cally acceptable sentences. Moreover, syntax was seen as the primary,
basic, or (in a technical sense) deep level of the language, with both seman-
tics (meaning) and phonology (sound structure) being constructed upon a
syntactic core. Thus, in a transformational account, the semantic and
phonological components have access to the output of the core syntactic
component, but not vice versa (Smith and Wilson 1979, p. 97): indeed, in
such an account, the verbal symbols are manipulated without any refer-
ence to their meanings or their sounds. Recently Chomsky (1981) has
nominated the shift from "language" to "grammar" in a restricted sense
as one of the two most important shifts in linguistic theory over the past
few decades (the other being the search for a more restricted notion of
grammars, so that possibilities for explanatory adequacy are enhanced).
Chomsky was also interested in language in a more abstract sense than
his predecessors had been. While the structuralists of the preceding gener-
ation examined what individuals actually said, and typically focused on
uttered words, Chomsky considered language to be an abstraction-a ca-
pacity that can merely be glimpsed in impure form in an individual's actual
output. Consistent with his formalistic leanings, Chomsky felt that lin-
guists should focus on this idealized, virtually Platonic form and disregard
the slight errors, pauses, lapses of memory, and the like which are as-
sociated with actual speaking in the real world. Both in his emphasis on
pure syntax and in his desire to study language as an ideal form, Chomsky
was dramatically at odds with the behaviorism of the time.
Verbal Misbehavior: The Controversy with Skinner

Chomsky gradually revealed his fundamental opposition to the com-
plete set of empiricist assumptions of most scientists, and nearly all lin-
guists, of his era. Initially, this revelation surfaced in a now notorious 1959
review of the book Verbal Behavior, by B. F. Skinner (1957), the arch-
behaviorist at mid-century. Skinner had attempted to explain linguistic
behavior-and the accent was on behavior-in terms of the same stimulus-
response chains and laws of reinforcement which he had invoked to ac-
count for the behavior of lower organisms-like the pecking of pigeons or
the maze running of rats. He had, for the most part, ignored the intricate
structural properties of language that fascinated Chomsky (and other lin-
guists) and other critics of behaviorism like Karl Lashley. And Skinner had
completely ignored the creative aspect of language-the fact that one is
191
free to talk about whatever one wants, in the way that one wants to.
Stressing the infinite expressive potential of language, Chomsky argued
that it is senseless to think of linguistic output as restricted in any mean-
ingful way by the stimuli that happen to be present.
With relish, Chomsky showed that Skinner's attempts to explain
language along these stimulus-response lines were fundamentally flawed.
They failed whether they were applied literally or metaphorically. As
Chomsky put it with characteristic asperity:
a critical account of his book must show that with a literal reading (where the
terms of the descriptive system have something like the technical meanings given
in Skinner's definitions) the book covers almost no aspects of linguistic behavior,
and that with a metaphoric reading, it is no more scientific than the traditional
approaches to this subject matter, and rarely as clear and careful. (1964a, p. 552)
But Chomsky made it clear that Skinner was also on the wrong track
epistemologically. Like other empiricists of the day, Skinner had urged
investigators to remain close to the data and to spurn abstract theory.
Chomsky, on the contrary, felt that the data would never speak for them-
selves, that it was necessary to take a theoretical stand and to explore the
consequences of that theory. Moreover, he revealed his own suspicion that
the kinds of theories needed to explain language, and other aspects of
human thought and behavior, would have to be abstract and, in fact,
frankly mentalistic. Leaning explicitly on the work of Descartes some three
hundred years earlier, and borrowing leaves as well from the writings of
Plato and Immanuel Kant, Chomsky argued that our interpretation of the
world is based on representational systems that derive from the structure
of the mind itself and do not mirror in any direct way the form of the
external world. And, indeed, once linguists accepted the impossibility of
a physical definition of grammaticality-for example, one expressed in
terms of actual utterances-they would realize the necessity for abstract
mentalistic linguistic theory.
In this book review, Chomsky also revealed his fundamental impa-
tience with most psychological approaches. In his view, psychological
experiments were often unnecessary to prove a point: many psychologists
preferred demonstrating the obvious to thinking systematically and in a
theoretical vein (Rieber 1983, p. 48). One psychologist Chomsky singled
out for praise was Karl Lashley. In his study of serially ordered behavior,
Lashley had concluded that an utterance is not simply produced by string-
ing together a sequence of responses under the control of outside stimula-
tion and interverbal association, and that the syntactic organization of an
utterance is not something directly represented in the physical structure of
the utterance. Rather, syntactic structure turns out to be, in Lashley's
192
words, "a generalized pattern imposed on the specific acts as they occur"
(quoted in Jeffress 1951, p. 122). Building on Lashley, Chomsky concluded
that there must be "a multiplicity of integrative processes which can only
be inferred from the final results of their activity" (1964a, p. 575).
A new psychological perspective appeared latent in Chomsky's for-
mulation. At the very end of the review, he opined, "At the moment the
question cannot be seriously posed, but in principle it may be possible to
study the problem of determining what the built-in structure of an infor-
mation-processing (hypothesis-forming) system must be to enable it to
arrive at the grammar of a language from the available data in the available
time" (1964a, p. 578). In the years ahead, Chomsky was to mine this
intuition, suggesting that the individual is born with a strong penchant to
learn language, and that the possible forms of the language which one can
learn are sharply limited by one's species membership with its peculiar
genetic inheritance. Chomsky was impressed by the extreme abstractness
of the task faced by every child who must learn language, the rapidity with
which the language is learned, despite the lack of explicit tutelage. Skin-
ner's notion of learning language through imitation and reinforcement was
implausible-if not impossible in the same sense as the finite-state gram-
mar: an entirely new psychological perspective was needed to account for
the rapidity and relatively error-free way in which children acquire lan-
guage, despite the "poverty of stimulus," the relatively small corpus of
often incomplete or error-tinged utterances they encounter in daily life.
General Messages: Chomskian Themafa

Chomsky hit his public stride in this essay. No longer simply a linguist
writing for a small group of colleagues about the arcane details of syntax,
he revealed himself to be a full-fledged cognitivist, one interested in many
dimensions of the mind and prepared to argue resolutely for his beliefs.
(The fact that Skinner never responded publicly to the review signaled to
many interested researchers the theoretical bankruptcy of the behaviorist
position.) And, in the years following the publication of his review,
Chomsky made clear others of his scientific themata. In his view,language
provided the best model for how to conceptualize and study thought
processes. Language should in fact be considered as part of a reformulated
psychology. Both language and the rest of psychology ought to exhibit
formal rigor, posit abstract models, and pursue the principles of explana-
tory adequacy. Informally phrased general principles (of the sort favored
by Skinner and many other psychologists) were anathema. Chomsky grad-
ually challenged the widespread belief in extremely general and broad
powers of the mind-powers like learning, stimulus generalization, and
193
the like. (In this move, he was reflecting, if not anticipating, similar shifts
occurring in psychology and artificial intelligence.) He came to think of the
mind as a series of relatively independent mental organs or modules, each
with its own rules of operation and development, along with prescribed
means of interacting with other "organs." Here was a powerful statement
on behalf of the autonomy both of language as an organ-and, not coinci-
dentally, of linguistics as a discipline. Superimposed on this modularity was
a commitment to mentalism-the existence of abstract structures in the
mind which make knowledge possible. And there was as well a swing to
nativism, the belief that much of our knowledge is universal and inborn, and
that, without the need for tutelage, one has access to this knowledge
simply by virtue of being human. These views were as much a departure
in the human sciences of the era as were Chomsky's initial claims in the
area of linguistics; and they were to generate even more controversy than
his initial views expressed in Syntactic Structures.
The full force of Chomsky's challenge to traditional linguistics be-
came clear in 1958 at the Third Texas Conference on Problems of Linguistic
Analysis of English. Here, in initially respectful but unfailingly firm terms,
he told his linguistic brethren that their traditional approaches to under-
standing language were doomed to fail. He asserted that a complete inven-
tory of elements in language could never give rise to a characterization of
all possible sentences, and that an inductive discovery procedure could
never work. He then went on to argue:
I think that the failure to offer a precise account of the notion "grammar" is
not just a superficial defect in linguistic theory that can be remedied by adding one
more definition. It seems to me that until this notion is clarified, no part of linguistic
theory can achieve anything like a satisfactory development. . . . I have been
discussing a grammar of a particular language here as analogous to a particular
scientific theory, dealing with its subject matter (the set of sentences of this lan-
guage) much as embryology or physics deals with its subject matter. (1964b, p. 213)
Chomsky took issue with the view that the methodological burden of
linguistics is the elaboration of techniques for discovering and classing
linguistic elements, and that grammars are inventories of these elements
and classes. Instead, he saw grammar as a theory of the sentences of the
language; and he saw the major methodological problem as the construc-
tion of a general theory of linguistic structure in which the properties of
grammars (and of the structural descriptions and linguistic levels derived
from them) are studied in an abstract way (Chomsky 1964b).
Following his presentation Chomsky engaged in debate from the floor
with leading structuralists of the past generation, whose views he was
opposing. While some of them had hoped to defeat the young upstart once
194
and for all, there was a very different outcome. As Newmeyer recounts on
the basis of the transcripts, "Here we can see linguistic history documented
as nowhere else-Chomsky, the enfant terrible, taking on some of the giants
of the field and making them look like rather confused students in a
beginning linguistics course" (1980, p. 35). In the years that followed, this
scene was to be repeated countless times, as Chomsky's steadily growing
body of admirers and followers did battle with the leaders of the once-
dominant structuralist school and demonstrated with numbing regularity
the inadequacy of earlier approaches and the promise of the new genera-
tive approach to grammar.
Chomsky's approach was indeed revolutionary-cataclysmic in rela-
tion to earlier linguistics; iconoclastic in relation to the prevailing empiri-
cist, inductivist, and nominalist temperament. Just as scholars like Levi-
Strauss in anthropology, Bruner and Miller in psychology, and Quine and
Putnam in philosophy had challenged the dogmas of their chosen disci-
pline, so Chomsky confronted received opinions in the linguistics world.
And because his approach was exceedingly formal, and the study of lin-
guistics relatively advanced in the human sciences, the scope of his claims
stood out with stark clarity.
Moreover, and this is part of the intellectual history of the time as
well, Chomsky's talents as a polemicist and his penchant for debate were
perhaps unmatched in recent scientific history. He liked to debate, was
brilliant at it, and was willing to argue tirelessly with all those interested
in his topic. He also inspired many students and colleagues to join him in
what seemed like a crusade-a crusade against the reflexive (and unreflec-
tive) empiricism of the previous generation, a crusade in favor of his views
of syntax, of language, of innate knowledge, of cognitive science, and, even
in many cases, of the contemporary political scene. In the end, many of
Chomsky's own supporters abandoned him, and today his particular lin-
guistic views are apparently held by only a minority of workers in syntax.
Yet even his most severe critics would not be studying what they are
studying, in the way that they are studying it, without his indispensable
and awesome example. And, indeed, both in the power of what he said and
in the uncompromising way he said it, Chomsky's work made a deep
impression from the first on nearly everyone who took the trouble to
study it.
All the same, as the excitement of those early days of transformational
grammar receded somewhat from view, it is easier to discern continuities
with earlier times. At the beginning of this chapter, I noted that Edward
Sapir, two generations older than Chomsky, was already dealing on an
intuitive level with many of the problems to which Syntactic Structures was
explicitly addressed. Going back much farther, the general epistemological
195
II I THE CoGNITivE SciENcEs : HISTORICAL PERSPECTIVE
approach that influenced Chomsky was taken from the work of Descartes,
and even more from those "Cartesian linguists" who were struck by the
creative powers of human language and who (like Wilhelm von Hum-
boldt) searched for ways to capture its regularities. And closer to home,
Chomsky shared the belief of his structuralist teachers in the importance
of studying syntax or grammar separately; the need (stressed by Roman
Jakobson and Louis Hjelmslev [1953]) for an essentially mathematical or
formal approach to language; the link (noted by Leonard Bloomfield) be-
tween the problem of describing a language and the problem of accounting
for language acquisition; and the conviction (with Ferdinand de Saussure)
that the study of langue (or competence) differs profoundly from the study
of parole (or performance). What Chomsky did, however, was to put these
various strands together: he showed that one could harness a formal-
logical analysis to illuminate intuitive relations among sentences, in service
of the goal of generating a system of rules which could account for all, but
only all, the correct sentences in the language. As Freud said of his discov-
ery of the unconscious through the "royal road" of dreams, such insight
falls to one's lot but once in a lifetime.
Linguistics of an Earlier Era
As far back as the classical era, scholars were arguing about the regularities
of language, and its ultimate characterization: is language part of nature or
part of culture? A system of grammatical categories, based on Greek and
Latin, evolved, and the study of rhetoric flourished as never since. Nor was
this interest exclusive to the Western world; in India in the fourth century
B.c., for example, Panini devised a grammar that was more diverse and in
some respects superior to that proposed by Greco-Roman scholars. During
the Renaissance, the tradition of classical philology reached great heights,
when it was applied to textual and critical analysis in various fields. As
scholars became aware of other languages, through travel and colonizing,
empirical knowledge accrued concerning the various languages of Asia,
Africa, Oceania, and the New World. There were even scientific expedi-
tions just to collect information about exotic languages (Robins 1967).
An important event in the history of linguistics occurred in 1786 when
Sir William Jones, an English orientalist, reported to the Bengal-Asiatic
Society that Sanskrit bore a striking resemblance to the languages of Greek
and Latin. Jones put forth the provocative hypothesis that all three lan-
guages must have come forth from a common source, which perhaps no
196
longer exists. This statement jostled the scholarly world, which had hith-
erto assumed that Latin had been a sort of corrupt Greek, and that most
European languages had just been derived from Latin. But a similarly
offhand explanation of the relationship between ancient languages and the
newly examined Sanskrit was not possible.
Jones's discovery ushered in a new conception of linguistic history,
linguistic phylogeny, and the processes of linguistic change. Earlier it had
been assumed that all languages had developed in the few thousand years
since Creation, and that Hebrew was the original language. Now instead,
linguists and philosophers became caught up with discovering the laws
that governed changes in language. Working in the manner of taxonomic
biologists, the linguists set up various language families (Semitic, Algon-
quin, Finno-Ugrian, and so on) and developed a theory of language
changes. Jakob Grimm found systematic correspondences between the
sounds of German and those of Greek, Latin, and Sanskrit. For example,
he noted that where Gothic (the oldest surviving Germanic language) had
an f, Latin, Greek, and Sanskrit frequently had a p. To account for these
relationships, he postulated a cyclical "soundshift" in the prehistory of
Germanic, in which the original voiceless (or unaspirated) stops became
"aspirates" (thus p becomes 1). But the systematization was not complete:
Grimm and his associates were willing to countenance exceptions to such
rules (Bolinger 1968).
The Neo-Grammarians
In the 1860s and 1870s, in a manner paralleling the Chomskian circle
a century later, a group of young rebels called the ]unggrammati/cer (or
neo-grammarians) attempted to put this situation in order. Not satisfied
with a mere collection or taxonomy of regularities, of the sort that Grimm
and his associates had noticed, the neo-grammarians claimed that the laws
of phonemic change admit no exceptions. Hermann Paul asserted in 1879,
"Every phonemic law operates with absolute necessity: it as little admits
of an exception as a chemical or physical law" (quoted in Brew 1968, p.
177}. Any apparent exceptions to these laws must be explained by refer-
ence to other laws, and there was little tolerance of theory for theory's
sake. As the neo-grammarians H. Osthoff and K. Brugmann declared:
Only that comparative linguist who forsakes the hypothesis-laden atmo-

sphere of the workshop in which lndogermanic rootforms are forged, and comes
out into the clear light of tangible present-day actuality in order to obtain from this
source information which vague theory cannot ever afford him, can arrive at a
correct presentation of the life and the transformations of linguistic forms. (Quoted
in Robins 1967, pp. 184-85)
197
The data-centered penchant of the neo-grammarians deeply affected

thinking about language. Clues to how language worked could be sought
in the physical characteristics of the utterances; and, in fact, the newly
devised techniques of psychophysics seemed promising tools. However,
despite efforts by the pioneering psychologist, Wilhelm Wundt, the psy-
chologists never really got a handle on linguistic phenomena. What is
interesting, in retrospect, is that at the tum of the century, the linguists
were monitoring psychologists, trying to figure out which of the new
psychological theories were correct; nowadays psychologists have been
dogging the linguists, attempting to determine which of the linguists'
perspectives is most likely to yield a psychologically valid view of language
(Blumenthal 1979).
de Saussure s Signal Contributions

While comparative philologists (like Grimm) enhanced empirical
knowledge about the development of language over the millennia, and the
relationships among the sound structures of various languages, their writ-
ings have a fundamentally dated quality: it is as if, in their proclivity for
discovering laws about the obscure past, these scholars had missed what
subsequent generations would see as the major point of linguistic study.
This was not true, however, of Ferdinand de Saussure, who can properly
be deemed the first linguist of the modem era. A Swiss savant, active at
the tum of the century, who made his first contributions to Indo-European
comparative linguistics, de Saussure (often abbreviated as Saussure) led his
colleagues away from their absorption with history and toward the inves-
tigation of the languages of their own time. He contrasted diachronic lin-
guistics (which focused on changes of language over time) with synchronic
linguistics (the study of language as a system at a single moment in time).
And he argued that much of importance in language could be understood
apart from those comparative and evolutionary concerns that had domi-
nated the work of his predecessors and his own early training.
Comparing language to a game of chess (an analogy of which he was
very fond), Saussure (1959) noted that, in the course of a game, the state
of the board is constantly changing: still, at any one time the state of the
game can be fully described in terms of the positions occupied by the
various pieces. On this view, it does not matter by which route the players
have arrived at their current positions. By a similar argument, though
languages are constantly changing, the successive or delimited states of
language can be described independently of one another and without
respect to their prior history.
Language, to Saussure, resembled a game of chess in still other ways.
First of all, the actual size and shape of the linguistic tokens is irrelevant:
198
just as any piece can stand for a bishop, any sound can stand for the
concept apple: all that is required is conventional agreement. Second, no
entity by itself has meaning: an entity accrues meaning only in terms of
its relationship to the other entities in an ensemble. Thus, the meaning of
a rook is only achieved in relation to what the other pieces can do in
principle and to what position they occupy at the present time; by the same
token, the meaning of a word (or a part of speech) is significant only in
terms of the other words in the language, and of the particular words with
which that given word happens to be concatenated at a specific moment
in time. Just as any move in a game of chess influences the complexion of
the game, and alters all the relations, so, too, any introduction of a new
word has reverberations across all other words in the utterance, text, or
entire language. And so language must be seen as an organized totality,
whose various parts are interdependent and derive their significance from
the system as a whole.
Saussure's focus on language as a system, and his interest in describing
that system in synchronic terms, were probably his most lasting contribu-
tions to the emerging science of language. (The Gestalt psychologists were,
at the same time, making similar points about the interrelationships among
visual perceptual elements.) However, other expeditions which he made
into the region of language were also telling. Saussure introduced a distinc-
tion between language (or langue) as a total system of regularities (like the
rules governing every game of chess) and speech (or parole) which are the
actual utterances themselves (like the moves of a particular game). While
parole constitutes the immediately accessible data, the linguist's proper
object is the langue of each community: the regularities in the lexicon,
grammar, and phonology which each person imbibes by being reared in a
particular speech community. Saussure shrewdly pointed out that a parrot
may speak the words but does not know the system of language; whereas
a person who knows the language of English may choose not to speak it.
(Shades of Searle's Chinese room conundrum!)
In some ways Saussure anticipated linguistic theory of today, and
several of his distinctions and conceptualizations are now part of the
working assumptions of every practicing linguist. For instance, he viewed
language as a cognitive system contained within the head of the individual
speaker-a mentalistic perspective that came perhaps readily to a French-
speaking scholar. But it is important to note differences between his out-
look and that of current workers. For one thing, Saussure was wedded to
a description or inventory of the elements in language and to building up
from the simplest to the most complex elements. This atomistic orientation
stands in sharp contrast to the Chomskian goal of a system of generative
rules which will yield only acceptable utterances. Moreover, Saussure paid
little attention to the properties of a sentence; in contrast, in Chomsky's
199
top-down approach, the rule-governed properties of the words in a sen-

tence and the relations that obtain among sentence types are of paramount
importance. Yet Saussure's concentration on the systematic and language-
specific properties of linguistics, and on the feasibility of explaining lan-
guage in its own terms, makes him a partner of contemporary writers. As
he insisted at the conclusion of his course, "From the incursions we have
made into the borderlands of our science, one lesson stands out. It is
wholly negative, but is all the more interesting because it agrees with the
fundamental idea of this course: the true and unique object of linguistics is language
studied in and for itself" (1959, p. 232). An important blow struck for the
autonomy view.
The Prague School

Having launched a science of linguistics, the Saussure school saw
centers rise around the world. Unguists in America, in France, and in
Scandinavia all acknowledged a debt to the master of Geneva. But the
Saussurean program was most successfully pursued by linguists working
in Prague in the 1920s and 1930s. Under its leaders Roman Jakobson and
Nikolay Troubetskoy, the Prague school exhibited wide interests but was
probably best known for its work on phonology. Rather than taking the
phoneme as the minimal unit of analysis, however, members of the Prague
school viewed phonemes as sets of distinctive features. Each phoneme was
composed of a number of articulatory features and was distinguished from
every other phoneme in the language by the presence or the absence of at
least one feature (Vachek 1966).
Jakobson explained the idea through a pithy example. Suppose one
meets someone at a party but does not get his name. It could be Mr. Ditter,
Mr. Bitter, Mr. Pitter, Mr. Titter, or the like. One's ability to figure out the
person's name depends upon hearing certain minimal contrasts. For exam-
ple, differ and pilfer differ on two concurrent minimal distinctions-which
happen to be called grave versus acute, and tense versus lax. Moreover, any
nonidentical set can be similarly described in terms of the features on
which they differ ..
Each of these distinctive features involves a choice between the two
terms of an opposition, a contrast that can be perceived by a listener.
Consider, for example, the sounds I pI and It I. These sounds feature a
contrast between grave and acute. In /pi there is a concentration of energy
at the lower end of the sound spectrum, whereas in It/ there is a concen-
tration of energy in the upper end of the spectrum. Thus the two sounds
have relatively different pitches, represent different distributions of energy
in production, and reflect resonating cavities with different sizes and
shapes. By the same token, all other distinctive features can also be defined
200
in terms of such perceptual, physical, and motor properties. These distinc-
tive features become aligned into simultaneous bundles of sounds which
are the phonemes of language. While one might think that there is a huge
multiplicity of features, Jakobson and his colleague Morris Halle reported
-again, in the significant year 1956-a mere twelve basic oppositions out
of which each language makes its own selections.
In its treatment of phonology along these lines, the Prague school
made its most characteristic and enduring contribution. Note that this
approach, while superficially similar to that of the descriptive structural-
ists, was actually motivated by different concerns. The Prague scholars
were interested in explanation, not just description, and had no hesitation
in imputing psychological reality to their linguistic accounts. Rather than
viewing the features as simply "convenient" ways of describing sounds,
members of the Prague school believed that they possessed physical and
psychological reality: the nervous system had evolved so as to ensure the
proper production and discrimination of these features. Features can be
defined independently of their appearance in any particular language; and
when found, they are accounted for by their place in a general theory of
phonology. Theory guides the discovery of phenomena rather than the
other way around.
The phoneme marked an important advance. First, it turned out that
the phonemic system of every language could be characterized in terms of
a small number of binary feature oppositions. Second, the features allowed
the formulation of generalizations that could not be stated in other struc-
turalist models. Also, the positing of features made possible the develop-
ment of an evaluation procedure: the optimal analysis should need the
least number of feature specifications per phoneme. These factors were
soon to exert influence on linguists who sought to bring comparable rigor
to the still fledgling study of syntax.
By no means did the Prague school restrict itself to phonology. Indeed,
Roman Jakobson and his colleagues were driven by a lust to write about
nearly every aspect of language, from phonology to poetry, and about a
large band of languages as well. Invading the field of psychology, Jakobson
(1941) sought to apply his interest in sound laws to a wide variety of
populations. He discerned a sequence of sound development in normal
children, one that follows what he called the law of maximal confrasf; and he
went on to demonstrate that this developmental progression is reversed in
an orderly way as a result of damage to the human brain. Jakobson
showed, furthermore, that the same kinds of contrast that had worked for
the description of phonology could also be applied to choices made at other
linguistic levels, ranging from syntax to pragmatics. And in a fertile metho-
dological contribution, Jakobson pointed out that most linguistic contrasts
are not equivalent or reversible: one pole of the contrast is more basic (or
201
unmarked), while the other is essentially defined in relation to the basic

pole, and is therefore marked. For example, at the level of sound structure,
a voiceless sound is unmarked in contrast to a marked one; at the level of
lexical semantics, the word long is unmarked, in contrast to the word
short which is marked. The concept of markedness found its way into descrip-
tions at each level of language and, as we shall see, exerted a major influ-
ence on adjacent fields.
By virtue of his knowledge of languages, his wide scope of interests,
and his boldness in theorizing, Jakobson inspired a whole generation of
linguists. And like other seminal scholars, he also exerted a pivotal effect
on other fields of knowledge. Probably his greatest impact occurred in
anthropology. Jakobson had met Claude Levi-Strauss in the early 1940s in
New York City, and the two of them had spent much time in consultation
and comradeship. At that time Levi-Strauss was searching for methods
that would aid in making sense of the masses of data about kinship and
social organization he had gathered during his field work in Brazil (Levi-
Strauss 1963). Introduced by Jakobson to the structural methods of linguis-
tic analysis, he seized upon them as a way of conceptualizing all cultural
data, not just those of language. For Levi-Strauss, as we shall see in chapter
8, the example of linguistics was the one to be followed by other, more
backward social sciences. But a vast problem inhered in deciding just how
to apply the phonological model in a region where it had not been explic-
itly formulated. In a more modest sense, this was the problem facing the
Prague school as well. One could readily use principles of analysis devel-
oped with reference to phonology, but just how to apply these principles
in more complex aspects of language, such as those involved in syntax or
poetry, was not always transparent.
Such reservations did not greatly trouble Jakobson. He was by tem-
perament a conquistador, not a doubter. Though chauvinistic about his
chosen discipline, he was not concerned to erect boundaries, to protect the
sanctity of phonology or linguistics from territorial snipers; in fact, he
strained to find connections to other fields. "I am a linguist/' he would
often declare "and nothing human is alien to me." Such forays and specu-
lative leaps were spurned by the strict conscience of the Chomskyites, and
for a time Jakobson's intellectual ambitions were seen as somewhat over-
arching and anachronistic. But now as the autonomy of linguistic study is
once more being challenged, his more synthetic viewpoint may regain
credibility.
Bloomfield Fashions a Field

Linguistics in America was also constructed in the shadow of Saus-
sure's program but, by virtue of the overwhelming influence of a single
202
linguist, Leonard Bloomfield, took a somewhat different historical tack
(Hockett 1968a ). From a contemporary cognitive perspective, Bloomfield
made two contributions-one wholly beneficent, the other far less so. His
positive work involved the development of methods and notations for the
study of unfamiliar languages. As a young instructor, Bloomfield began to
study various exotic languages-for example, Tagalog, the principal In-
donesian language of the Philippines. Bloomfield worked from dictation,
taking down an extended discourse, which he then subjected to detailed
analysis. Rejecting the model of Latin grammar, which had guided many
of his predecessors, Bloomfield recorded the sound patterns of each partic-
ular language, noting both its similarities and its departures from other
classical languages. From the study of Tagalog, Bloomfield moved to Ma-
laya-Polynesian and Algonquin. In each case he developed appropriate
methods for recording the language and struggled to find the best way of
describing its sound patterns and grammatical regularities. The identifica-
tion of constituent structures, as a means of syntactic analysis, is an
outgrowth of Bloomfield's work.
In mastering the properties of these various languages, Bloomfield was
not content simply to pursue his own research. He came to see the need
for a cadre of trained linguists. Soon he founded a linguistics society and
launched a series of publications. He saw linguistics as a natural science
-one patterned after physics:
As the physicist need not follow the path of each particle, but observes their
resultant action in the mass, and their individual actions only when these in tum
group themselves into a deflection of the mass condition (as in radio-active sub-
stances) and rarely has occasion to watch the impingement of a single particle, so
in linguistics we rarely attend to the single utterance or speaker, but attend to the
deviations of utterances and speakers only when they mass themselves into a
deflection of the total activity. (Bloomfield 1925, p. 2)
In campaigning for the discipline of linguistics, and in modeling how

to record new languages, Bloomfield set an enviable standard. Like Wundt,
he deserves credit for founding the principal institutions in his field. Yet,
again like his German counterpart, Bloomfield's substantive vision had its
narrow aspects and, in the latter's case, ended as counterproductive for the
further development of his science.
In the early 1920s, Bloomfield fell under the influence of A. P. Weiss,
a behaviorist at Ohio State, and gradually embraced a completely behavio-
rist view of language. The view evolved by Bloomfield became physicalist
(and hence anti-mentalist): he thought of all language as a purely physical
phenomenon and rejected any special psychological or mentalistic features
in the explanation of linguistic behavior. Thus, for instance, while he
would have liked to be able to account for meaning, Bloomfield came to the
203
conclusion that it was dangerous to deal with this concept. He declared

that, in order to give an accurate definition of meaning, one would need
a scientifically complete knowledge of everything in the speaker's world.
This being impossible, Bloomfield preferred to speak in simple behavioral
terms of the recurrent features of a situation in which a certain form is used
(Newmeyer 1980, p. 9).
Ultimately he reached a most conservative conclusion: "The only
useful generalizations about language are inductive generalizations. Fea-
tures which we think ought to be universal may be absent from the very
next language that becomes accessible" (quoted in Newmeyer 1980, p. 4).
He declared his stand in favor of strict behaviorism, mechanism, and
operationalism. One only studies elements that are overtly visible or tangi-
ble, lend themselves to mechanistic explanations, and will ultimately yield
stable predictions. He inveighed against mentalism, declaring that it would
be as thoroughly discredited as Ptolemaic astronomy had been. Indeed, in
the effort to avoid mentalistic explanations, he restricted his gaze to "pri-
vate ... events of physiology," on the one hand, and to the external stimuli
of objects or people, on the other.
By dint of his clear precepts and his forceful personality, Bloomfield
dominated the American linguistic scene for most of the first half of the
present century. His Language (second edition, 1933) was for many years the
standard text: all students of linguistics were familiar with, even if they
did not wholly endorse, the behaviorist and antisemantic tone of the
volume. It was against the ghost of Bloomfield, and such lively and loyal
followers as Charles Fries (1963) and Charles Hockett (1958), that
Chomsky and his colleagues launched their transformational revolution.
But if Bloomfield dominated the linguistics scene, he did not wholly
control it. Indeed, Edward Sapir, who had been trained by the pioneering
anthropologist Franz Boas, put forth a set of views that were to intrigue
anthropologists as much as Bloomfield's views captivated the main line of
linguists. Sapir did not avert meaning: indeed, he devoted some of his most
probing writings to this troublesome but essential component of language.
And he put forth the provocative hypothesis that a person's very processes
of thinking are structured, if not controlled, by the particular properties of
the language one speaks. As he expressed it in his textbook Language:
Human beings do not live in the object world alone, nor alone in the world
of social activity as ordinarily understood, but are very much at the mercy of the
particular language which has become the medium of expression for their society.
It is quite an illusion to imagine that one adjusts to reality essentially without the
use of language and that language is merely an incidental means of solving specific
problems of communication or reflection. The fact of the matter is that the "real
world" is to a large extent unconsciously built up on the language habits of the
204
group .... We see and hear and otherwise experience very largely as we do because
the language habits of our community predispose certain choices of interpretation.
(1921, p. 75)
In the hands of his student Benjamin Lee Wharf, Sapir's ideas were
further developed. Wharf did not shrink from the conclusion that even our
most basic notions might be derived from language. While Newtonians
might think that space, time, and matter reflect the same intuitions every-
where, Wharf demurred. He claimed that the Hopi Indians have totally
different conceptions of these Kantian categories, and that these distinc-
tions have grown out of the different ways in which their language parses
the universe. As he put it:
We dissect nature along lines laid down by our native languages .... The
world is presented in a kaleidoscopic flux of impressions which has to be organized
by our minds-and this means largely by the linguistic systems in our minds.
. . . No individual is free to describe nature with absolute impartiality but is
constrained to certain modes of interpretation even while he thinks himself most
free. (1956, pp. 153, 213-14)
Many linguists were intrigued by the work of Wharf and Sapir, even
as many psychologists and anthropologists found it an excellent source of
fertile hypotheses. (In chapter 12, I shall examine some of the cognitive
scientific work it inspired.) And yet, like much of the work in semantics
of the period, it was considered peripheral to the mainstream of linguists
-too difficult to lay one's hand on, too much contaminated by the "real
world" and by the dangerous notion of "meaning," not susceptible to
sufficient documentation or proof. Indeed, Sapir and Bloomfield stood in
marked contrast to one another. Bloomfield assumed the pose of a rigor-
ously scientific worker, concentrating on methodology and formal analy-
sis. His insistence on a strictly mechanistic approach to meanings and his
pessimistic attitude toward semantics contributed to the relative neglect of
that subject matter. Far more reminiscent of Jakobson, Sapir ranged widely
and was not afraid to touch on culture, conceptualization, and the world
of meaning. Paradoxically, he shared many of Chomsky's intuitions about
linguistic structure but little of the formalistic inclinations of either Bloom-
field or Chomsky. A union of Bloomfield-Sapir approaches and insights
has yet to take place in linguistics.
The Crisis of the Early 1950s

When Bloomfield died in 1949, from most perspectives the discipline
of linguistics appeared in fine shape in the United States. Commentators
205
spoke of "great progress," "far-reaching advances," "a flourishing state,"

and definitive results." The first half of the century had seen the conflu-
I/
ence of historical, comparative, and structural descriptions of language; the

discovery and development of the phonemic principle; and the attempts
to put grammar and other aspects of a descriptive analysis on a firm
footing. Phonology served as a model for grammar, semantics was seen as
separate from grammar, and language was viewed as inherently system-
atic. Inductive discovery procedures were endorsed, and the assumption
prevailed that one could gradually build up from the simplest level-that
of sound-to word level (morphology), to syntax, and thence to semantics.
While not all the decisive steps had yet been taken, it was widely assumed
to be only a matter of time before the various forms and levels of language
were completely understood (Newmeyer 1980).
Adding to this sense of progress and accomplishment were the rapid
advances in allied fields. Theories of information (dealing with channels
of communication) and techniques like stochastic modeling (dealing with
probability) seemed like natural allies for the analysis of language. The
invention of the sound spectrograph for the analysis of speech samples
promised to lay bare the physical properties of the linguistic signal. As
historian Newmeyer puts it, "many linguists felt that the procedures had
been so well worked out that computers could take over the drudgery of
linguistic analysis. All one would have to do (in principle) would be to
punch the data into the computer and out would come the grammar!"
(1980, p. 2).
Yet, on closer analysis, the scene was not as auspicious as most com-
mentators of the time thought. First of all, as I have noted in earlier
chapters, the hegemony expressed by empiricist philosophy and behavio-
rist psychology was breaking down rapidly. As these views began to fall
into disrepute, the basis on which Bloomfield's linguistic platform had
been built was undercut. Even scholars not influenced by this philosophi-
cal revolution had begun to raise doubts about the methods used by
linguists. It was not as easy as it seemed simply to collect the data and to
describe them objectively: one's own expectations, one's own knowledge
of meaning, inevitably colored the data and lent an unsettling circularity
to apparently objective descriptions. Even Hockett, a defender of the con-
temporary approach, declared in 1955 that it was not possible to analyze
a language objectively-one had to "empathize" with the information
(quoted in Newmeyer 1980, p. 17).
Efforts to repair the house of structuralist linguistics occurred from
within. Probably the most successful attempts were those of Zellig S.
Harris (1952) who met difficulties in a purely descriptive approach to
syntax by positing transformations-rules that make explicit the intuitive
206
relations among certain sentences. But while the techniques Harris in-
vented were subtle and genuinely useful for organizing the data, he re-
sisted strenuously an attempt to tie all the transformations into a single
theoretical framework, or to posit an abstract level of analysis. Chomsky
learned about these methods from his teacher Harris; and, as has been
suggested by several commentators, it was inevitable that someone of a
more theoretical bent than Harris would try to draw out the implications
of such a transformational approach for the contemporary efforts to write
grammars. The result, as we have seen, was Chomsky's Syntactic Struc-
tures, a work in which he posited a level of analysis more abstract than the
surface sentence, on which the transformation(s) would be imposed. More
generally, rather than simply noting possible transformations between
sentences, Chomsky posited a system of rules that map from abstract
underlying structures onto the articulated sentences of language.
The Evolution of Chomsky's Thought
Period Pieces
It is not possible in a survey to detail the particular models, with their
numerous alterations, put forth by Chomsky and his followers in the years
following the publication of Syntactic Structures. Nonetheless, it is important
to stress that Chomsky's contribution did not occur at a single moment in
time and does not consist of a single body of theory. Indeed, few of the
specific notions put forth in Syntactic Structures have remained in their origi-
nal form (even though the overall program has been maintained with
considerable fidelity). Kernel sentences were soon dropped; the number of
transformations was reduced and simplified enormously; the relation
among the levels of syntax, semantics, and phonology has come to be
understood along somewhat different lines. Certain topics introduced in
that initial volume have been abandoned; while others, merely hinted at
in 1957, have spawned entire research programs.
The first major new statement came in Chomsky's monograph Aspects
of the Theory of Syntax (1965). This book presented what is now termed the
11
Standard Theory," the theory that for many years was the staple of
introductory textbooks in linguistics. In the Standard Theory, there are no
longer initial kernel sentences. Instead, one now starts with the base gram-
mar, which generates an initial phrase marker, the deep structure. Major
operations are performed upon this deep structure. There is a transforma-
207
tional component which converts the initial deep structure into other
structures, the final of which is the surface structure; in this theory, most of
the transformations are obligatory.
Two interpretive components operate on these core syntactic compo-
nents. Deep-structure relations are interpreted by a semantic component:
thus, the information necessary for semantic analysis must be represented
in the deep structure. Phonological interpretation occurs on the surface-
structure string: thus, the initial deep syntactic structure is "read" for
meaning, while the ultimate surface structure is "read" for sound. In the
1965 version, both the base and the transformation rules were uncon-
strained, permitting a large variety of possible base and transformational
systems.
The Standard Theory was a more ambitious theory, attempting in part
to accomplish for semantics what had been modeled for syntax alone in
Syntactic Structures. It also proved far more controversial and, because of
various insufficiencies, eventually had to be abandoned. Particularly con-
troversial was the notion of deep structure. Deep structure is a technical term,
having nothing to do with profundity. Nonetheless, it was frequently
misinterpreted in this fashion. Moreover, even among those who under-
stood its technical scope, deep structure was seen as "too deep" by some
critics and as "not deep enough" by others. The move of making semantics
an interpretation of core syntactic arrangements was also widely debated.
We can distinguish two broad lines of attack on Chomsky. One line
of attack came from the conservative branch of linguistics-from scholars
still attached to the structuralist perspective of the Bloomfield days.
Charles F. Hockett selected himself to answer Chomsky on behalf of the
traditional interest groups in structural linguistics. In The State of the Art
(1968b ), Hockett argued that the fundamental program of transformational
linguistics was flawed. Chomsky had bet on the wrong scientific model. He
had compared language to a formal discipline like logic or mathematics;
whereas it is an empirical science like chemistry. Only humanly invented
systems like logic have the regularity Chomsky sought in language; in
contrast, natural language is ill defined, and the theories of computability
and algebraic grammar he sought to apply are irrelevant. "This knocks the
props out from under current mathematical linguistics, at least in the form
of algebraic grammar, whose basic assumption is that language can be
viewed as a well-defined subset of the set of all finite strings over a
well-defined finite alphabet" (1968b, p. 61). Hockett also rejected the
separability of grammar and semantics (which he had earlier espoused) and
the separation of both from the rest of culture. Autonomy within linguis-
tics, and from other disciplines, was abandoned. In his view, the "grammar
of language" in Chomsky's sense simply does not exist.
208
What is striking was the absence of response to this monograph.
Except for a dismissive review by Chomsky's one-time associate George
Lakoff (1969), this work of criticism was ignored. Apparently by the late
1960s, ripostes from the earlier generation had simply become irrelevant.
In accordance with Thomas Kuhn's (1970) description of paradigm changes
in the sciences, the Chomsky point of view took over, not by convincing
the previous generation that it had been in error, but by winning the
allegiance of the most gifted students of the succeeding generation.
Some members of the younger generation felt, however, that
Chomsky was not sufficiently radical. In the late 1960s a major attack on
his work was launched by certain linguists who had been sympathetic to
his general position. Led by Chomsky's former students, George Lakoff
and}. R. Ross (1976), these critics questioned the simple positing of two
levels of analysis-a deep structure, to be interpreted semantically, and a
surface structure to be interpreted phonetically-and also called into
doubt the autonomy of syntax from semantics. Ultimately, these critics
abandoned simple deep structure in favor of grammars whose underlying
structures were much deeper and closer to semantic representation them-
selves. (Another group of critics, which eventually included Chomsky
himself, embraced grammars in which the underlying structures were
shallower and closer to surface structure than had previously been en-
visaged, and were equipped with much richer lexical components than
had earlier been the case.) The generative semanticists contended that
there was no dear-cut distinction between syntactic and semantic rules,
and that a level of syntactic deep structure defined as the initial genera-
ting component could not be sustained. Instead, they set up rules that
take semantic representations as their input and yield surface structures
as their output, with no intervening level of deep structure. In contrast,
the interpretive semanticists, including Chomsky, transferred more of the
work of syntax into the semantic component, so that deep structure has
gradually moved closer to the surface structure of a sentence (Smith and
Wilson 1979).
The debates between the generative semanticists and the traditional
Chomskians, and, eventually, between the generative semanticists and the
interpretive semanticists, were ardent and often vicious. At stake were
particular claims about how best to represent word meanings, as well as
competing views about how to construct linguistic theories. Names were
called and epithets were hurled in a manner that shocked even seasoned
polemicists. When all the hue and cry had died down, however, the gener-
ative semanticist school lost its steam, and most of its principal adherents
abandoned the field of syntax altogether. Meanwhile, in his persistent
way, Chomsky has continued to investigate syntax. It is true, however,
209
that the problems pointed out by the generative semanticists have stimu-
lated the Chomsky contingent to develop certain notational forms (using
symbolic logic) as well as a plethora of new and highly abstract mech-
anisms, all designed to deal with inadequacies in the traditional transfor-
mational approach. Perhaps most decisively, generative semantics helped
to sound the death knell for the transformational component; this once
central component has been entirely eliminated on some accounts or radi-
cally simplified on others.
Since the late 1960s, Chomsky's brand of generative grammar (as it came
to be called) has undergone a series of changes. It is not possible to intro-
duce and follow the intricacies of the extended standard theory or of its
current replacement government and binding theory, let alone the specifics
of trace theory or X-bar theory-the major landmarks in the development
of his approach up to the present. I can, however, indicate some of the
broad trends in the theory. Over the years, Chomsky has steadily nar-
rowed the definition of his object of study. Never interested in language
as an overall communication system, he now questions whether language
per se is a system worth trying to study at all. Viewing language as a more
abstract notion than grammar, more remote from actual mechanisms,
Chomsky is more firmly convinced than ever that linguists should concen-
trate on solving the issues of syntax (1982, p. 14). There are no more
attempts to systematize semantics.
Responding to many criticisms, Chomsky has attempted to reduce the
expressive power of transformational structures. It turns out that the origi-
nal grammars had been so powerful that they could not illuminate the
crucial issue of what is permissible in a human language (see Peters and
Ritchie 1973). If one wanted to know which grammars might characterize
human performance, it was necessary to restrict what a candidate grammar
could do. This attempt to weaken transformational power has been part
of what Chomsky has called the second major shift in linguistics-the
search for principles that would constrain the set of syntactic rules of
natural language.
Chomsky has tried to reduce the class of transformations by discover-
ing general conditions that rules must meet: he has sought to eliminate the
possibility of compounding elementary operations to form more complex
transformational rules. As he now puts it:
[M]uch effort has been devoted to showing that the class of possible transfor-
mations can be substantially reduced without loss of descriptive power through the
discovery of quite general conditions that all such rules and the representations
they operate on and form must meet .... [The] transformational rules, at least for
a substantial core grammar, can be reduced to the single rule, 'Move alpha" (that
is, "move any category anywhere"). (Mehler, Walker, and Garrett 1982, p. 21)
210
It may even be possible to eliminate the phrase-structure rules wholly
or in large part. As transformational rules and phrase-structure rules be-
come less prominent, more attention is being paid to the lexicon-to the
particular rules governing specific words. The lexicon now contains much
of the information that used to be part of the transformational apparatus.
And the notion of surface structure (now sometimes reconceptualized as
5-strucfure) becomes much richer. In fact Chomsky has even commented
that his new theory could be considered as a "one level theory," consisting
merely of base generation of S-structures (see comment in Longuet-
Higgins et al. 1981, p. 279).
How then, to characterize Chomsky's current approach to language
or, more strictly speaking, to grammar? He describes his pursuit in theo-
retical terms as a search for universal grammar (or U.G.). U.G. is genetically
determined at its initial state (in the organism) and is specified, shar-
pened, articulated, and refined under conditions set by experience, to
yield the particular grammars found in specific groups of individuals (1980,
p. 234). A theory of universal grammar is said to be an explanatory the-
ory. To know a language is to be in a certain state of mind/brain: this
state is described by a core grammar which consists of certain principles
of universal grammar, which are to be discovered by linguists. Thus,
Chomsky takes a purely realistic stance: language knowledge is a series
of states in the brain.
What are these principles to which Chomsky refers? Here we come
to what he considers to be another major shift in his theory. Until fairly
recently, his goal had been to describe the various rules that individuals
must somehow know if they are to know a language. But now he has come
to the conclusion that it is more productive to speak of various princi-
ples that govern language use. These principles begin in human biology.
They determine the kinds of grammar that are available in principle
(Chomsky 1979). The principles of U.G. are various subsystems, which go
by names like binding theory, control theory, government theory, theta theory, and the
like-each of which features a limited degree of parametric variation.
If U.G. is sufficiently rich, then even limited linguistic data in the
environment should suffice for developing rich and complex linguistic
systems in the mind. There is no need to talk of the acquisition of rules.
Rather, each of the systems of U.G. has certain parameters associated with
it, and these are set or fixed in light of the data encountered by a person
(ordinarily a young child) in the course of acquiring his native language.
Slight changes in the values of the parameters proliferate throughout the
system to yield what on the surface may be rather different language
structures (Chomsky 1982}. The grammar of a particular language that
ultimately emerges can be regarded as a set of values for each of these
211
I I 1 THE CoGNITIVE SciENCES : HISTORICAL PERSPECTIVE
parameters: the overall set of rules, principles, and the like constitute U.G.,
part of the human language faculty.
We see here the union of two visions in the science of the mind. One
vision, stemming from philosophy and dating back to Plato, features lan-
guage as a kind of idealized object, governed by a small set of universal
principles, having relatively few parameters (Chomsky 1981). The other
vision, coming from biology, puts forth language as an organic system, or
module, which has the potential to develop in a small and delimited range
of ways; the particular path of development yielding a "core grammar" is
determined by the kinds of information encountered by the organism.
Embracing these two visions, one can account at once both for the similari-
ties across all languages (thanks to universal grammar) and the distinct
differences among particular languages (thanks to the variations in param-
eter setting).
What is one to make of such a complex theory which has undergone
critical changes in the course of a few years and will doubtless change
further? One approach, sometimes adopted by Chomsky's former stu-
dents, is to stress the differences in the theory from one period to another
and to attribute these differences to efforts to shore it up in the light of
competing accounts. Such a critical point of view has been put forth by
George Lakoff (1980). In the view of this former Chomsky student,
Chomsky had to eliminate deep structure and transformations because it
had proved impossible to maintain the principal assumption of modularity
-the independence of syntax. Once Lakoff and his associates had shown
that meaning and use affect virtually every rule of syntax, Chomsky had
to narrow progressively the domain of syntax: rules once regarded as
clearly within the domain of syntax were redefined as part of semantics.
On this account, Chomsky has conceded the validity of criticisms in prac-
tice while, in his explicit remarks, denying that he has done so. The
historian of American linguistics, Frederic Newmeyer, offers another inter-
pretation of this phenomenon. In his view each generation of Chomsky
students continues to work on the major ideas proposed by their teacher
during their own intellectually formative years:
Chomsky in the early 1960's was an"abstract syntactician"-and many of his

students from that period still are! Chomsky in the late 1960's proposed the lexical-
ist alternative to abstract syntax-and those who were then his students are still
developing this model. And we can predict that Chomsky's 1970's students will
be refining trace theory long after Chomsky has developed his ideas along other
lines. (1980, p. 207)
In my own view, Chomsky has certainly been influenced by the

criticisms leveled at him, both by former students and from others who
212
have some general sympathy with generative grammar. However, he
does not like to concede this influence explicitly. He is not easy to dis-
pute with, and does not suffer criticism gladly, and so tends to be dismis-
sive in writing even (or perhaps especially) of those scholars whose work
has had some influence on him. This rhetorical style tends to polarize
discussion.
Despite undeniable shifts in emphasis and strategy, however, it is
remarkable how Chomsky has adhered to the program he initially pre-
sented to the scholarly community in Syntactic Structures and had been pur-
suing from his early twenties. The centrality of syntax, the belief in a
transformational component, the view of semantics as an interpretation of
basic syntactic relations-all these have endured.
Chomsky is correct in saying:
My major interest has been to make precise the basic principles which enter
into the knowledge of language that has been attained by the speaker-hearer; and
beyond that, to try to discover the general theoretical principles which account for
the fact that this system of knowledge, rather than something else, develops in the
mind when a person is placed in a certain linguistic environment. In a general way,
I might say that I am still working very much within the framework of ... early
unpublished work. (1979, p. 113)
Any growing science will be constantly changing. Chomsky has been one
of the principal revisers of his own theory, sometimes in radical directions,
even though the same vision has guided him from the start and he has been
relentless in pursuing it.
As he moves to increasingly abstract characterizations, involving tech-
nical argumentation, he has indeed lost adherents. This does not seem to
bother Chomsky, who (like many another revolutionary figure) has always
seen himself as somewhat of a loner. As he declared in a recent interview:
The particular domain into which I put most of my energies, the structure of
language, seems to me to have been a very exciting one just in the last seven or
eight years. I don't pretend to speak for any consensus in the field here, in fact,
I'm in a very small minority in the field in this respect, but I believe it's been
possible in the past few years to develop a theory of languages with a degree of
deductive structure that provides a kind of unification and explanatory power
going well beyond anything that would have been imagined even a decade ago.
Again, I don't think many linguists would agree with me about this .... I suppose
I'm in a very small minority in the field today. But then, that has always been the
case. With regard to me, it doesn't seem very different now from what it was ten
or twenty years ago. But my own views are not what they were then, and I hope
they will not be the same ten years from now. Any person who hopes to be part
of an active growing field will take that for granted. (Quoted in Rieber 1983, pp.
62-63)
213
II I THE CoGNITIVE SciENCES : HISToRICAL PERSPECTIVE
Reactions in Other Cognitive Sdences
By the early 1960s, when Chomsky was gaining widespread support

in linguistics, his work was also coming to the attention of workers in other
fields. Since cognitive science has placed its trust in such cross-disciplinary
exchanges, it is important to consider the course of interaction between
contemporary linguistics and other cognitive fields.
One of the very first scholars outside linguistics proper to become
aware of Chomsky's work was the psychologist George Miller. By the
early 1960s, Miller (1962) had become a convert to Chomskian linguistics
and soon helped to tum the psychology of language into a testing ground
for Chomsky's transformational claims. Miller and his students tried to
figure out ways in which to demonstrate the "psychological reality" of
transformations: they hypothesized that the steps by which a sentence is
theoretically generated and transformed are also realized by the "live"
individual in the process of comprehending or producing sentences. This
effort was not particularly successful, but important methods of psy-
cholinguistic research were worked out in the process (Fodor, Bever, and
Garrett 1974).
Miller and his colleagues succeeded in bringing Chomsky's work to
the attention of the psychological community. Chomsky got a fair hearing,
and a minority of students fully bought his program for language and
cognitive science. Still, at this writing, the majority of workers in psychol-
ogy have remained skeptical about the overall relevance of his theory for
their pursuits. Although Chomsky himself describes linguistics as part of
psychology, his ideas and definitions clash with established truth in psy-
chology. He has had to contend not only with the strong residue of behavi-
orist and empiricist sentiment but also with suspicion about his formal
methods, opposition to his ideas about language as a separate realm, and
outright skepticism with respect to his belief in innate ideas. While
Chomsky has rarely been defeated in argument on his own ground (for a
recent dramatic example, see his debate with Piaget), his particular notions
and biases have thus far had only modest impact in mainstream psychol-
ogy (Piattelli-Palmarini 1980).
There is one area of psychology where Chomsky's ideas and example
have had enormous influence: the psychology of language, or psycholinguis-
fics, as it is often called. In the past few decades, psycholinguistics has be-
come a major area of inquiry, encompassing studies of normal language in the
adult, the development of linguistic capacities in children, the breakdown
of language after conditions of brain damage, and the use of language in ex-
ceptional populations such as the deaf. A great deal of work in each of these
areas has focused on syntactic capacities, with models for analysis generally
supplied by Chomsky. At times these models have been used as a means of
214
characterizing the data collected; at other times, the data have been used to
test the "psychological reality" of models-the extent to which linguistic
behavior unfolds according to the principles put forth by Chomsky.
After two decades of psycholinguistic work, most psychologists have
despaired of applying Chomsky's work in any direct way to their own
research. Not only are Chomsky's formulations highly abstract and subject
to frequent change (typically in a more abstract direction); but when they
have been applied, the results have not been consistent with his models,
at least in any straightforward way. For example, any number of develop-
mental psycholinguists have sought to account for language acquisition in
terms of Chomsky's categories and derivations, but these efforts have
generally been judged failures. Recently there have been interesting at-
tempts to develop learnabilify theory-a formal account of the constraints
that must be built into a child's cognitive apparatus if one is to learn
language from the data to which one is exposed (Wexler 1982; Wexler and
Culicover 1980). Steven Pinker (1984) has attempted to interpret the data
of language acquisition in terms of principles of leamability. It is too early
to say whether this approach will signal a new use for a Chomsky-inspired
(or "generative") approach within psycholinguistics or whether a Chom-
skian perspective will continue to be a minority taste.
For the most part, then, Chomsky has been more influential in psy-
cholinguistics because of the kinds of question to which he has drawn
attention than because of any direct utility of his theory for experimenta-
tion. Sometimes he and his followers have discounted empirical research
in psycholinguistics, with the disclaimer that their theories have to do with
idealized competence, and not with the facts of individual performance. To
my mind, this is an unjustified maneuver. Chomsky and his followers are
only too happy to cite empirical data when it appears to accord with their
theory. However, it is often far from clear just how directly Chomskian
ideas are meant to be translated into empirical work; and, in that sense at
least, the Chomskian reservation is justified.
Chomsky's new ideas were introduced to the philosophical world
primarily through the work of two young Princeton graduates, Jerrold Katz
and Jerry Fodor, who were teaching at Massachusetts Institute of Technol-
ogy in the early 1960s. Collaborators for several years, Katz and Fodor
(1963) also developed a model of semantics which became incorporated
into the "standard version" of transformational grammar. As has been the
case in psychology, Chomsky's ideas about language have persuaded a
minority of younger philosophers, but many others are decidedly ambiva-
lent about his specific claims. The formalism itself presents less of a prob-
lem, but a number of the core ideas have had a checkered history in
philosophy. Philosophers have reacted coolly to Chomsky's promotion of
seemingly discredited rationalist notions and to his enthusiasm for innate
215
ideas. His ready use of terms like rules, s!rudures, systems, with (apparent)
disregard for the nontrivial technical problems involved in such concepts,
and his facile reinterpretation of leading philosophical figures of the past
(Descartes as a hero, the empiricists as villains) have proven difficult for
most philosophers to swallow. Also Chomsky's lack of interest in seman-
tics has troubled many philosophers, who find in the work of semanticist
Richard Montague some of the same formal elegance others have admired
in Chomsky's syntactic discussions (Thomason 1974).
Yet it is fair to say that, as in psychology, topics of discussion have
been materially affected by the fact that Chomsky has spoken and written.
Whether the impact of Chomsky's notions will be greater in a half-century
than they are now, or whether they will be seen as a curious aberration
within the general triumph of empirically oriented and anti-mentalistic
sciences is too early to say. Except in the field of linguistics itself, it remains
uncertain whether Chomsky's ideas will emerge as germinal and essential.
And what of other cognitive sciences? While his approach arose from
many of the same roots as artificial intelligence, several of Chomsky's main
ideas are not readily implemented in computational formats. For example,
there is no guarantee in principle that one can parse sentences using trans-
formational grammatical approaches. Moreover, A.I. is very much oriented
toward practical problems of designing programs that understand sen-
tences or stories, and Chomsky's syntax-centered framework is not suited
for the main issues of understanding discourse. Accordingly, computer
scientists like Roger Schank have been publicly hostile to the theory,
taking the position that semantics and pragmatics are central in language
and that syntax is relatively unimportant. Schank has also attacked the
modular notion: "It is impossible to produce a model of language alone
... apart from beliefs, goals, points of view and world knowledge" (1980,
p. 36). Other scholars, like Terry Winograd, have borrowed some ideas
from transformational grammar but have not made Chomskian theory
pivotal to their systems. Efforts to parse sentences using transformational
ideas have been few thus far (though see Berwick and Weinberg 1983;
Marcus 1980). Even through Chomsky's formal elegance has appealed to
A. I. researchers, his Platonic view is even more remote from most computer
scientists than from the average psychologist and philosopher. For his part,
Chomsky has been rather critical of research in artificial intelligence,
finding it mostly unmotivated and ad hoc: he does, however, admire David
Marr's work on vision (1982).
Rival Positions within Linguistics

Those linguists who continue to work in the generative grammar
tradition, but who have broken from Chomsky's particular perspective,
216
have sometimes been attracted to the artificial-intelligence perspective.
One such researcher is Joan Bresnan, formerly a colleague of Chomsky's
at M.I.T., who has concentrated on developing a theory of language which
is psychologically real (1978, 1981). In opposition to Chomsky, who (as I
noted) pays little attention to how his derivations might be realized by an
11
individual speaker operating under real-world" constraints, Bresnan and
her colleagues have fashioned a perspective designed to illuminate how an
individual will perceive or produce .language. In her lexical-functional the-
ory, there is no transformational component. The information traditionally
embedded in the syntactic components is now placed in the individual's
lexicon-one's knowledge of specific words. Lexical-functional grammar
provides each sentence with two structures: a constituent structure, which is
similar to a surface structure (or phrase-marker tree) in the standard theory
of Chomsky; and a functional structure, which includes all the grammatical
relations relevant to the semantic interpretation of the sentence. The func-
tional structure is generated by annotated phrase-structure rules work-
ing in conjunction with lexical entries for the various morphemes in a
sentence.
In recent years, Bresnan has begun to work closely with colleagues in
the area of artificial intelligence and in psycholinguistics. The purpose of
this collaboration is to determine whether the modifications she has intro-
duced into standard syntactic theory make her position "psychologically
viable" in a way Chomsky's has never been. The results thus far suggest
that the kinds of parsing mechanisms devised in the light of her theory do
comport better with experimental data on language processing and with
the models of understanding being developed by workers in computa-
tional linguistics. For similar reasons, in his work on learnability theory,
Steven Pinker (1984) has embraced lexical-functional grammar, which he
views as a "central tendency" (or modal position) among contemporary
linguistic theories. The work of Bresnan and her colleagues has also be-
come central in the new, Stanford-based Center for the Study of Language
and Information, in which the avowed goal is to achieve an integrated
theory of human linguistic competence. The Center has just been
launched, and it is too early to say whether its dream of uniting the
philosophical, psychological, and computational aspects of language will
be met; but all of us involved in the cognitive science movement will be
monitoring progress there with abiding interest.
Much of the appeal of Bresnan's theory also accrues to another point
of view which has recently gained adherents-the theory of Gerald Gazdar
of the University of Sussex. Reverting to the generalized phrase-structure
grammars that Chomsky strongly attacked in his early publications, Gaz-
dar (1981) argues that one does not need transformations, and that even
unusual surface structures can be stated in a straightforward way; more-
217
over, it is important to do this because sentences thought to be the same

by transformational grammarians are often actually different from one
another. In Gazdar's theory, semantic interpretation is applied directly to
the surface structure generated by a grammar. There are explicit semantic
rules for each syntactical rule. Gazdar believes that his theories are more
straightforward than Chomsky's; comport better with analyses done in
formal language theory (which, in tum, are relevant to the devising of
language systems on computers); and provide a more appropriate entry to
problems of language acquisition. Carving out territory in opposition to
Chomsky, he asserts:
When one realizes that the syntactic theory that Chomsky has been develop-
ing over the last ten years has embraced phrase structure rules, complex symbols,
a level of S-structure, a level of D-structure, a level of "Logical Form," filters,
transformations, interpretive rules, stylistic rules, coindexing conventions, and
abstract cases, among other things, it is a little surprising to hear him castigating
as "needlessly complex" an alternative syntactic theory that employs only phrase
structure rules, complex symbols, and a level of surface structure. (Quoted in
Longuet-Higgins, Lyons, and Broadbent 1981, p. 281)
To readers not intimately involved in linguistics, the differences between

Gazdar and Bresnan are not easy to comprehend, and even their diver-
gences from Chomsky may seem relatively modest. (Chomsky himself
tends to minimize these differences, often terming the different theories
"notational variants" of the same core ideas.) Yet inasmuch as these theo-
ries are rivals in the search for programs or mechanisms that can "truly
comprehend," it is important to determine which model is most appropri-
ate. Current supporters of both Bresnan and Gazdar are working at the new
Center for the Study of Language and Information, and they expect to
determine whether the Bresnan or the Gazdar model is more appropriate
for handling linguistic data or whether (as Chomsky might predict) the
differences between the positions prove minor. What might constitute a
crucial test of the theories, however, is not apparent. Linguistics is not (yet)
mathematics or physics.
A Tentative Evaluation
While it is far from clear whether Chomsky's particular positions will

ultimately prevail-either in linguistics proper or in neighboring disci-
plines-he has framed the issues for debate. His notions of which issues
218
Linguistics: The Search /or Autonomy
are important in linguistics continue to dominate discussions in much of
the profession: and his ways of formulating issues have influenced all
cognitive scientists, even scholars overtly inimical to his work and young
students only dimly aware of the source of their views.
If the index of importance of a scholar is the extent to which one could
be replaced within a discipline, Chomsky clearly emerges as the most
important figure in the linguistics of recent times-equal in importance to
de Saussure and to Jakobson in earlier eras and one who may ultimately
exert an even greater influence on the direction of his field. Within the
broader cognitive sciences, his contribution is more controversial and less
secure. But on the bases of his demonstrations of the mathematical preci-
sion implicit in language, his modeling of the importance of theory-driven
research, and his potent arguments for mentalism, nativism, and mod-
ularity, I would offer my opinion that he is one of the two or three most
important and least replaceable thinkers of the whole movement. Piaget
may have turned up a greater number of important phenomena; Herbert
Simon and Allen Newell may have put forth a paradigm more widely
emulated by other investigators; but no one has framed the issues of a
cognitive science with as much precision and conviction as has Chomsky.
Moreover, while not even he has fully lived up to the rigorous criteria he
has demanded of a linguistic theory, the criteria he devised in his early
works continue to be those by which subsequent linguistic theories (and
theorists) are judged. Whether his impact will be as broad in other cogni-
tive sciences may depend upon whether the model, which has proved
fertile in the area of language, proves equally useful in other areas, such
as visual perception, logical classification, or the study of consciousness.
Chomsky himself might feel of two minds about this. On the one
hand, he is possessed of a rigorous scientific conscience and would like to
see his degree of formal precision and his criteria for explanatory adequacy
invoked everywhere. On the other hand, he has long insisted that the rules
governing language may be unique to that domain, and he is loath to
endorse notions of general cognitive structures that cut across diverse
contents. And so, while the study of linguistics in which he has pioneered
may serve as a model of how one goes about investigating other fields, it
remains an open question whether any parallels of substance will emerge.
This paradoxical situation is epitomized by Chomsky's view of lin-
guistics as a part of psychology. In repeatedly giving voice to this senti-
ment, Chomsky may appear to be suggesting that the study of language
ought simply to be incorporated into a more general study of psychology
(a catholic view). Yet it seems clear to me that Chomsky would not ap-
prove of incorporating linguistics into psychology as currently practiced.
Psychology would need to be reconfigured in Chomskian terms (a far
219
narrower point of view which might find little sympathy in psychology

circles).
What, then, about the claims for autonomy in the area of language?
Chomsky has for almost thirty years insisted that syntax can be ap-
proached as a module, one operating in independence of (though, of
course, interacting with) other facets of language. While many researchers
have sought to follow Chomsky in this belief, it now appears that the links
of syntax to other aspects of language-to lexical semantics (a Ia Bresnan)
and to pragmatics (a Ia Gazdar) in particular-may be sufficiently integral
that the thesis of the autonomy of syntax is in jeopardy.
The point is not whether syntax can be looked at separately-of
course, any component can be examined in isolation. The question is,
rather, what is to be gained or lost by adhering rigorously to such a
research program. In calling the autonomy of syntax to account, critics
argue that the most important operations of language are better conceived
along different lines. They call for consideration of models where the
interactions among various factors are brought to the foreground (as, say,
in lexical-functional accounts) in place of a model where efforts are di-
rected chiefly at determining the operations of syntax in isolation, and
where the relation between syntactic and other components is relegated to
a later phase of research.
In short, then, Chomsky's demonstration of autonomous syntax is still
important on methodological grounds, but it is less evident that it can be
considered a viable characterization of the central aspects of language.
Instead, collaborative approaches in computational linguistics-where the
pragmatic and communicative aspects of language are frankly conceded
and the interactions between semantic and syntactic factors are presup-
posed-may carry the day.
And what of the relation of language study to other cognitive
sciences? One tack is to bring other disciplines to bear on language, while
continuing to treat language as a domain apart, worthy of investigating on
its merits. Here language-the province of linguistics-is illuminated by
other disciplines-as happens in neurolinguistic studies of speech produc-
tion or in psycholinguistic investigations of phoneme recognition. Many
scholars (including the original advisers to the Sloan Foundation) have felt
that language is the best testing ground for an integrated cognitive science.
There could then be parallel studies of an interdisciplinary type directed
at other domains of knowledge, such as vision, motion, and motor action.
This is dearly a model for cognitive science-one designed along "vertical"
lines, and one for which I have much sympathy.
The degree to which the study of language ought to remain separate
from other scholarly disciplines depends to some extent on the issues one
220
is most interested in illuminating. If one is interested in language as an
abstract system-be it the creation of some divine force or simply a pattern
emerging from the brain-then language is appropriately studied in terms
of the kinds of taxonomic and structural category favored by linguists.
(Language here becomes a "distanced" object of study, analogous to the
solar system probed by astronomers.) If, however, one is interested in
language as it participates in human intercourse, then a view of linguistics
as divorced from other disciplinary pursuits becomes less tenable.
Consistent with this latter point of view, some scholars have come to
adopt (or to readopt) a more horizontal model of cognitive science. On this
view, it is wrong to think of the subject matter of any discipline (for
example, language as the subject matter of linguistics) as privileged: in-
deed, it is more important to connect any human activity to the range of
related fields that can be investigated. Furthermore, language does not
belong to any discipline but is instead a part of every cognitive scientific
discipline, efforts to cordon it off being artificial or wrong-headed. This
was the faith that influenced scholars like Jakobson in the area of poetics,
Sapir in his studies of language and thought, and de Saussure in his
historical studies of language. Calling for a closer integration of fields, Roy
Harris, a harsh critic of Chomsky's, declares:
Language cannot be studied in isolation from the investigation of "rational-

ity." It cannot afford to neglect our everyday assumptions concerning the total
behavior of a reasonable person .... An integrationallinguistics must recognize that
human beings inhabit a communicational space which is not neatly compartmen-
talized into language and non-language .... It renounces in advance the possibility
of setting up systems of forms and meanings which will "account for" a central
core of linguistic behavior irrespective of the situation and communicational pur-
poses involved. (1981, p. 165)
Chomsky (1980) has sometimes conceded that language may be less

of a "cordoned off" territory than he would like, and that the demands of
communication or the intrusion of belief structure may suffuse all linguis-
tic activity. Should this be the case, however, language would not be
worthy of study because it would prove to be hopelessly intermeshed with
everything else. In Chomsky's words:
If non-linguistic factors must be included in grammar: beliefs, attitudes, etc.

[this would] amount to a rejection of the initial idealization to language as an object
of study. A priori such a move cannot be ruled out, but it must be empirically
motivated. If it proves to be correct, I would conclude that language is a chaos that
is not worth studying ... Note that the question is not whether beliefs or attitudes,
and so on, play a role in linguistic behavior and linguistic judgements ... [but
rather] whether distinct cognitive structures can be identified, which interact in the
221
II I THE CoGNITIVE SciENcEs : HisTORICAL PERSPECTIVE
real use of language and linguistic judgements, the grammatical system being one
of these. (1979, pp. 140, 152-3)
We confront here the most basic issues of scientific strategy. It may

be that, in the last analysis, Chomsky is wrong. Perhaps the connections
between syntax and semantics, between the "language system" and other
mental organs, or between linguistics and other disciplines, is stronger or
differently configured than he had imagined. It may, nonetheless, be the
case that Chomsky has selected the optimal research strategy. For even if,
ultimately, everything turns out to be connected to everything else, a
research program rooted in that realization might well collapse of its own
weight.
One of the most attractive features of cognitive science is its positing
of methods and models that are sufficiently rigorous to allow the analyst
to determine where those models are insufficient or unsatisfactory. The
computational paradox in fact arises because computational methods have
helped us to recognize some of the ways in which humans deviate from
a simple-minded logic machine. Chomsky's work is exemplary in this
regard because he has fashioned rigorous methods for syntactic analysis.
Whatever role syntax ultimately turns out to play in human cognition, its
mechanisms are beginning to be understood, and its relations to other
components of language are beginning to be clarified. This remark can be
made about few other areas in the cognitive sciences. For providing a
paradigmatic example of clear-cut scientific progress in this new field,
Chomsky has earned the respect of cognitive scientists everywhere.
As exchanges continue to be conducted with other cognitive scien-
tists, it will become easier to assess the overall impact-and the ultimate
limitations-of Chomsky's single-minded vision. How language fits into
the wider culture-a topic ruled off limits (for methodological reasons) by
Chomsky-will certainly be an issue addressed in a broad field. Of particu-
lar import in determining the limits of a Chomskian approach will be
intense collaboration between linguistics and anthropology, two fields
whose histories have commingled from the very beginning. In just what
ways these connections were originally forged, and how they are manifest
today, will become clear as I consider anthropology as a cognitive science.
222
8
Anthropology: Beyond
the Individual Case
Lucien Levy-Bruhl Examines the Mind of the Primitive
Initially, Lucien Levy-Bruhl had seemed a progressive voice in the long-

standing debate about the rationality of the savage mind. Yet, during his
career, the sentiments among his anthropological colleagues changed to
such an extent that ultimately he was perceived as being conservative,
almost reactionary, in his views. Finally, by the end of his life, the French
thinker had undergone an almost complete change of mind: he renounced
the very position that had initially drawn attention to his work. In examin-
ing closely the reasons for these shifts, we not only gain a better under-
standing of a paradigmatic figure in the history of anthropology, but also
confront the central enigma of cognitively oriented anthropology.
When Levy-Bruhl began his study of the thinking processes of primi-
tive peoples almost a century ago, the general orientation in the anthropo-
logical community was evolutionary. It was assumed that members of
"advanced" Western civilization represented the height of reasoning, and
that "lesser individuals" around the world were simply inferior copies of
the Western mind. Levy-Bruhl challenged this received opinion: "Let us
then no longer ... want to reduce their mental activity to an inferior form
of our own." Primitives do not reason badly; rather, they reason differ-
ently. Levy-Bruhl proposed that the primitive mind follows a kind of logic,
a "pre-logic," which is fundamentally different from our own: these pre-
logical thoughts had best be understood on their own terms (quoted in
Cazeneuve 1972, p. 41).
223
Levy-Bruhl proposed two major characteristics of primitive thought.

First of all, such thought partakes of the law of parficipafion. Primitive in-
dividuals see objects, beings, and other external phenomena as at once
identical with themselves and also as something other than themselves.
Thus, the Trumai (a tribe in north Brazil) say that they are aquatic animals,
meaning not that they are only fish, but rather that they are both human
beings and fish at the same time. Second, and in a fashion no less incom-
prehensible to the Western mind, primitive individuals can emit andre-
ceive mystical forces, properties, and qualities, which are felt as outside
themselves, without their ceasing to be what they are. Thus, a newborn
child can suffer the consequences of everything his father does, eats, says,
and the like (Cazeneuve 1972). Given these properties, the primitive mind
does not abstain from contradiction. There is no need to give up an idea
because it collides with another one: both notions can be entertained
simultaneously.
In positing different ways of thinking, Levy-Bruhl pointed up the
problems involved in extrapolating from one human population to an-
other. There is no substitute for a careful study of the "mental" representa-
tions of each group. As a result of such study, he concluded, the mind of
the primitive emerges as quite different from that of contemporary Euro-
peans: dominated by emotion and affect, little concerned with logic, toler-
ant of contradictions and mystical forces that would be taboo in a civilized
Cartesian context.
Scholars typically defend their views when attacked by the succeed-
ing generation. But, at least privately, Levy-Bruhl violated this pattern.
Indeed, in the papers of his posthumously published notebooks, Levy-
Bruhl engaged in a debate that was as severe and tortured as any in
professional publications. He became dubious that primitives really do
exhibit a different, pre-logical form of thought:
The step which I have just taken, and hope is decisive, consists, in a word,
in abandoning a badly posed problem ... even allowing for the numerous and
characteristic cases of participation of which my six volumes are full, there still exist
doubts about the explanation . . . . I started by positing a primitive mentality
different from ours ... a position which I have never been able to defend well, and
in the long run an untenable one.... The thesis thus extenuated and weakened
is no more defensible . . . . Let us entirely give up explaining participation by
something peculiar to the human mind. . . . There is not a primitive mentality
distinguishable from the other. (Quoted in Cazeneuve 1972, pp. 86-87)
In these later musings, Levy-Bruhl virtually adopted the position he

had originally opposed. He concurred with those who see differences be-
tween primitive and civilized as a question of degree, and was no longer
224
Anthropology: Beyond the Individual Case
remote from those who doubted there was any fundamental difference
whatsoever. Levy-Bruhl took refuge in the argument that, perhaps, it was
scientific thought that is unusual: in the paraphrase of Jean Cazeneuve,
"What has been described to us under the name of primitive mentality is
undoubtedly a permanent structure of the human mind but in our society
this structure is blurred by the supremacy of scientific thought whereas it
remains in the foreground among preliterate peoples" (1972, p. 22). He
concluded with a sentiment that would gain widespread endorsement
today: "The fundamental structure of the human mind is the same every-
where" (p. 12).
The debate that raged in Levy-Bruhl's mind echoed a discussion that
has been widespread in the West ever since humans first became aware of
the existence of exotic populations and began to ponder their relation to
more familiar pockets of humanity. Though the debate extended to the
morals and values of these alien populations, keen interest always centered
on the quality of their thought processes. Did they have the same logics
as "we" do, or were there logics peculiar to each group of individuals? How
could we get inside the mind of the primitives, and discern the world in
the way they see it? Those scholars who acquired a professional interest
in these questions were "at risk" to become anthropologists.
In the time of Levy-Bruhl, early in this century, the sources of evi-
dence were principally textual: one read the myths or, less frequently,
transcripts of conversations with preliterate individuals, and then drew
conclusions about the kinds of thought reflected in them. But such her-
meneutic methods could not attain scientific status. And so, increasingly
in the twentieth century, anthropologically oriented scholars repaired to
the field to examine firsthand the thought processes of "their people." This
move to "case studies" was certainly an important step in an empirical
direction. The problem with individual fieldwork, however, was that it left
a great deal of discretion-perhaps too much-in the hands of a single
investigator or a small cadre of fieldworkers.
A significant chapter of the history of anthropology in this century
has been a search for methods of inquiry that were less idiosyncratic, more
reliable (Herskovits 1953; Kardiner and Preble 1961). Since it was generally
not practical for large teams of investigators to visit the same site-and if
investigators visited at widely disparate times, they might not be witness-
ing the "same" peoples (Freeman 1983)-a premium was placed on more
objective methods which could be employed by a single investigative
team. This need gave rise in the 1960s to the field of e!hnoscience, a seemingly
objective set of empirical measures by which one could assess the thinking
processes of peoples everywhere.
Ethnoscientists had no problem in invoking a representational level.
225
What for other more experimentally oriented disciplines constituted a

major leap came naturally to anthropologists. Problems lay in another
direction. While it seemed for a while that the new empirical procedures
might place anthropology on a firmer scientific footing, there has recently
been a disaffection with these methods: however elegant, they seem to
bypass too much of what is central in the thinking and the experience of
a culture. There has been at least a partial return to the view that anthro-
pology ought to re-embrace the holistic methods of the in-depth case
study, and perhaps align itself more with the humanities and less with the
sciences. Continuing uncertainty about whether anthropological investi-
gation ought to adopt the methods of experimental cognitive science con-
stitutes a second major theme of the discipline.
Speculations about different populations date back to classical times.
At least as far back as Herodotus, it was known that human behavioral
patterns differed from one society to another, and attempts were made to
document and explain (or explain away) such differences. Students of the
Bible often traced the existence of exotic peoples back to some episode-
for example, the dispersion of the three sons of Noah, or the offspring of
the banished Cain, whose face had been covered with darkness. Quite
often, it was assumed that certain exotic populations, like American Indi-
ans or Arabian nomads, represented a corruption or degeneration of a
"pure" human: they stood in contrast to a strain that had descended
directly from Adam or Abraham, and was, of course, related to contempo-
rary Europeans.
With the revived interest in scholarship, and the far-flung travel that
characterized the Renaissance, such age-old speculations gained new force.
Some commentators continued to hold to the view of degeneracy; others
stressed resemblances between the contemporary savage and the predeces-
sors of civilized man. When comparisons were made between the early
Europeans and contemporary savages, the implication followed that con-
temporary savages were simply at an earlier stage of development than
contemporary Europeans. As one commentator has put it, "Europe and
America ... settled the matter by placing its [sic] own nations on one end
of the social series, and the savage tribes at the other, arranging the rest
of mankind between these limits according as they correspond more
closely to savage or cultural life" (Hodgen 1964, p. 483).
Some scholars simply sought to understand the differences. Exem-
plary in this regard was the French savant Charles Louis Montesquieu,
who concentrated on the study of laws, customs, and morals from different
parts of the world. In an effort to explain why societies differ from one
another, Montesquieu postulated factors like population density, geo-
graphical barriers, degree of isolation, stages of technological development,
226
subsistence patterns, state of commerce, climate, and soil. In his view,
natural surrounds and conditions of climate played a more decisive role in
savage societies than in more developed societies. As an early functionalist,
he sought within each society a motivated basis for such seemingly brutal
practices as cannibalism, slavery, or idol worship (De Waal Malefijt 1974).
The advent of the Enlightenment, with its ardent belief in rationality
and equality, sharpened the debate about the mentality of the primitive.
On the one hand, to the extent that rationality was the standard, the
Western mind seemed qualitatively more advanced than that of the appar-
ently confused savage. On the other hand, to the extent that equality was
stressed, another set of conclusions seemed warranted. Such practices as
slavery, and such beliefs as the superiority of one group over another, were
seen as regressive or anachronistic. These emerging egalitarian points of
view posed difficulties for those of a religious persuasion, who needed to
explain why some groups believed in a single God, while others clung to
polytheistic notions. The notorious Bishop Whately claimed that savages
could not be helped and were best thought of as members of a different
species. Darwin posed a different threat to members of established
churches who now had to contend with his demonstrations that all hu-
mans were descended from a line of forerunners, dating back millions of
years, and that human beings could not be thought of apart from the rest
of the Natural Order (De Waal Malefijt 1974).
Edward Tylor Launches the Discipline of Anthropology
The stage was set for more systematic thinking about different human
groups. Various scholars helped to initiate the scientific study of society
and culture, but the person most often granted this honor (or responsibil-
ity) is an Englishman of the late nineteenth century, Edward Tylor. Work-
ing at the same time as Wilhelm Wundt, the founder of modern psychol-
ogy, Gottlob Frege, the inventor of modern logic, the neo-grammarians in
linguistics, and the first generation of experimental neuroscientists, Tylor
published his magnum opus, Primitive Culture, in 1871. Having toured
America and Mexico in the 1860s and obtained thereby a vivid notion of
cultural differences, Tylor undertook in his book a rationalist assault on
the divine inspiration of religious beliefs. According to his revisionist
perspective, human culture and religions were products of a natural, law-
governed evolution of human mental capacities.
A new field requires definitions, and Tylor produced the most-often
227
quoted description of culture: "That complex whole which includes

;knowledge, belief, art, morals, law, custom, and any other capabilities and
· habits acquired by man as a member of society" (1871, p. 1). The term
acquired was critical: Tylor was declaring that human capacities are not
simply part of one's birthright: they are rather derived from one's member-
ship in a group and presumably could be changed, if the individuals were
reared in a different group or if the group itself altered its practices or its
values. This emphasis on learning also undercut any notion that individu-
als behave as they do because of particular inheri~ed characteristics or their
niche along an evolutionary scale. If behavior can be learned or acquired,
it can also be altered.
Tylor was clearly under the sway of Darwinian evolutionary ideas.
According to his own scheme, humanity could be arrayed along a linear
track, ranging (in his terminology) from savagery, to barbarism, to civiliza-
tion. He believed in psychic unity, however, and held that all peoples were
capable of making this progression. Further, even those individuals at the
height of civilization were not bereft of earlier traces. According to Tylor's
scheme of survivals, every individual possesses many habits, beliefs, and
customs that date back to earlier times and have endured despite their
current lack of utility. As a timely example of a survival, the phrase "God
bless you" long since has lost its original meaning and yet is ritualistically
invoked when someone sneezes. Conversely, Tylor also held that even the
most irrational customs are products of a reasoning capacity like our own;
like Montesquieu, he believed that if one only understood the origins and
circumstances of a custom, it would make sense.
In addition to contributing new ways of thinking about culture and
about the relationships among different groups, Tylor also made important
methodological contributions to the science he was helping to found.
Noteworthy was his method of adhesion, whereby he attempted to deter-
mine which customs or practices hang together. This he accomplished by
preparing massive lists of the practices carried out in various cultures, and
noting which tended to occur at the same time. By such correlation, he
could show, for instance, a predictable relation between teknonymy (nam-
ing one's parents after children) and living in the house of the mother's
relatives; moreover, this practice of naming parents after children proved
even more closely related to practices of avoidance, where individuals with
a potentially tense relation assume a protective distance from one another.
This method of correlation was a signal contribution to the new field.
According to Robert Lowie, an anthropologist active in the first half of the
twentieth century, "Nothing that Tylor ever did serves so decisively to lift
him above the throng of his fellow-workers" (quoted in Kardiner and
Preble 1961, p. 75). Instead of metaphysical speculations about why a
228
certain practice came about, Tylor could now ferret out statistical relations
between different institutions or practices.
From our contemporary perspective, Tylor's views may seem to blend
with those predecessors who saw the primitive as just a pale version of
civilized modem man. In fact, however, Tylor helped to undermine this
view. To argue that culture actually exists among all men, in however
crude or primitive a form, was a major step toward a more relativistic point
of view. Moreover, Tylor's belief that all groups harbor vestiges of the
past, and that behavior can be understood and justified if seen in context,
served to bring the primitive person closer to the circle of civilized modem
man (Stocking 1982). As Abram Kardiner and Edward Preble have com-
mented, to overcome the ecclesiastics "who would create an impassable
gulf between civilized man and his primitive ancestors, Tylor had to show
that the 'rude savage' was potentially an English gentleman" (1961, p. 77).
Could Shaw's Pygmalion be far behind?
The British Scene
Tylor's work on primitive societies links the armchair characterizations of

earlier centuries with the empirically oriented work of the present era: the
course from speculation to correlation. Following his pathbreaking work
in England, one could witness both of these forces struggling for the
ascendancy. Representing the old guard was the student of ancient soci-
eties, Sir James Frazer, whose multivolumed The Golden Bough (originally
published in 1890) may well be the most famous anthropological book ever
written. In this beautifully wrought work, which influenced many human-
ists even as it mesmerized the general reading public, Frazer traced a
connecting thread from the pagan ceremonies of the past to the practices
of Christianity and other modem religions. He described early forms of
magic where one could control another individual simply by gaining pos-
session of some vestige of that individual. These totemic practices an-
ticipated the rise of religion where individuals gave up the belief that they
themselves could control events, and instead posited nonhuman higher
powers which govern the world. And finally, Frazer described the highest
stage of development, that of science, where man once again began to
manipulate nature, but this time sought to uncover and test the relevant
physical laws. On this view, early men and contemporary primitives were
seen as relatively irrational, though perhaps possessing the same potentials
as modem man (Frazer 1955).
229
Though Frazer's work was much admired by nonscientists, and read

by most anthropologists, the tradition he represented eventually yielded
to a less grandiose, more empirical approach. An event that symbolized this
trend was the launching, just before the turn of the century, of a large-scale
expedition to the Torres Straits in the South Pacific. Never before had so
many men of a scientific stripe embarked on a mission simply to document
what life was like among a very different group of people. Proceeding
(paradoxically) at Frazer's suggestion, A. C. Haddon, once a zoologist,
organized the Cambridge Anthropological Expedition. Haddon was inter-
ested in primitive mentality but, unlike his predecessors, proposed to take
systematic measurements of psychological characteristics in the field. Ac-
companying Haddon were experts in psychology, medicine, linguistics,
and music, including W. H. Rivers, C. S. Myers, and William McDougall
-all psychologists with medical training who would later become major
social scientists (Rivers 1900).
The scientists on the expedition carried out many investigations. Es-
pecially when viewed from our current perspective, they did not focus on
"higher" cognitive functions; instead, following the practices of psychia-
trists of that era, they probed abilities to make discriminations in various
sensory modalities, to appreciate illusions, and to name colors. Nor were
the results particularly decisive with respect to the controversy about
primitive mentality. There were some provocative findings: for example,
a hint that the language available to individuals might influence the way
in which they see or group colors; documentation of the Papuan's keen
powers of observation (in the face of unremarkable visual acuity); a sug-
gestion that the perception of spatial relations may also be culturally
conditioned; and the documentation of capacious memories for family
genealogies. McDougall also claimed that the sense of touch of natives was
twice as delicate as that of Englishmen, while sensitivity to pain was only
half as great. In general, however, the scientists reached no consensus on
the scope of differences between the groups nor, in the case of the differ-
ences, on which reasons would best account for them (R. Brown 1956;
Stocking 1982).
But the Torres expedition was a landmark in the history of anthropol-
ogy, not because of its results, but for having taken place at all. That six
major scientists could spend several years in the field, carrying out careful
observational and experimental studies, was a dramatic demonstration that
cultural differences need no longer be simply speculated about: they could
be examined critically. It took some years before the full significance of this
model had been absorbed by the rest of the scientific community. But, in
the end, the kind of speculative statements offered by Frazer simply could
not compete with empirical findings "in the field."
230
The American Version
Boas s Scholarly Hegemony

Thanks to the unstinting efforts of a recently transplanted German
physicist, Franz Boas, the same lessons were being driven home in the
United States (Herskovits 1953; T. Kroeber 1970; Stocking 1982). Boas had
originally come to the New World to execute his doctoral research-an
improbable study of the color of sea water. He had become dissatisfied
with the laboratory study and wanted to study the color of sea water
firsthand-or so goes the legend. Boas found himself in the Arctic area,
near Baffinland, carrying out geographical studies, when he first encoun-
tered Eskimos and became intrigued by their languages and their behavior.
Thus arose a lifelong fascination with natives in this part of the world and
a long-term commitment, by a gifted and energetic man, to the founding
of a scientific anthropology (as he declared) "before it was too late."
Faced with the conflicting claims of the physicist-who sought objec-
tive explanations of color-and the explorer-who sought to capture the
atmosphere of exotic cultures--Boas strove to reconcile these perspectives.
He concluded that validity must be granted both to the scientific view of
the outsider and to the subjective view of the particular individual or
culture. Here, then, was Boas's resolution of the "universal" versus "par-
ticular" dilemma which confronts all anthropological workers. He brought
this lesson to the larger arena of anthropology, where he undertook a
long-term study of Indian societies in the Pacific Northwest. In addition,
he began to train nearly all of the next generation of anthropologists,
including not only those who remained faithful to his credo, but even
those who went on to found rival schools.
Apart from Boas's total commitment to the calling of anthropology,
it is not simple to summarize his contribution: He steadfastly avoided
strong theoretical statements, preferring to adopt a more inductive ap-
proach. Still, one can readily point out the directions in which he led
anthropology, directions that remain discernible even in today's complex
disciplinary terrain.
First of all, Boas opposed the notion of the linear evolution of culture.
Taking issue with Tylor and those who had gone before him, Boas felt that
each culture was best studied in terms of its own practices, needs, and
pressures, rather than in relation to some other culture which represented
a more or less advanced mode of organization. In his view, one should
carefully document what is done and try to understand why, rather than
231
evaluate how advanced or simple it might be (as the evolutionists insisted

in doing), or where it came from (as the diffusionists were wont to do). This
focus on the well-documented individual case, along with an abjuring of
evolutionist or diffusionist temptations, has characterized anthropology in
the post-Boas era.
Boas emphasized the importance of language and of linguistics for all
of anthropological study. Gifted at learning languages, he developed meth-
ods for the careful notating of languages and stimulated many of his
students to document the Indian languages while they were still in use.
The pioneering American linguist Leonard Bloomfield paid tribute to
Boas's remarkable achievements in this area:
His greatest contribution to science, and, at any rate, the one we can best
appreciate, was the development of descriptive language study. The native lan-
guages of our country had been studied by some very gifted men, but none had
succeeded in putting this study on a scientific basis .... Boas amassed a tremendous
body of observation, including much carefully recorded text, and forged, almost
single-handed, the tools of phonetic and structural description. (1943, p. 198)
Boas saw each language as a unique creation which needed to be

understood (in Saussurean fashion) as an organized totality. He empha-
sized that languages could converge from different sources, as well as
diverge from a common source, and deplored the nineteenth-century
search for a basic, or "Ur,"language from which all other languages sprang.
He also underlined the important role of language in all of human activity,
though he expressed skepticism that a culture could be restricted by the
form of its particular language. Indeed, in opposition to the Whorf-Sapir
perspective, Boas saw thought as influencing language rather than vice
versa. Even if some languages are less given to abstract terms than others,
all are sufficiently complex to carry out the varieties of human thought.
Thanks to Boas's insistent message, there has always been a tie in Ameri-
can scholarly circles between the fields of anthropology and linguistics-
a model of the type of cooperation toward which cognitive science now
strives.
Boas's thoughts about language related closely to his ideas about
primitive mentality. In his view, the principal difference between the men-
tal processes of primitive peoples and ourselves inheres in one fact:
whereas the categories used by the primitive have developed in a crude and
unreflective manner, contemporary literate populations have been able to
systematize knowledge, in the manner of the rational scientist. This differ-
ence has emerged not because each individual in our society thinks in a
more logical manner but rather because various philosophically oriented
materials have become worked out more systematically over the genera-
232
tions and are now available to the general population. Boas's own explora-
tions had convinced him that Indians and Eskimos could appreciate ab-
stract language and thought if confronted with it, but that this set of
concerns was simply not part of their habitual intellectual milieu. Overall,
then, Boas was one of the first, and one of the strongest, advocates of the
view that primitive and modem individuals possess essentially the same
cognitive potential; an enduring suspicion of unwarranted or undocu-
mented dichotomies between "primitive" and "modem" is one of his most
compelling contributions to contemporary anthropological thinking.
From the start of his career, Boas had been suspicious of any attempt
to evaluate one human being, or one human group, as better or worse than
another; as his career thrived, he used his increasingly powerful and in-
creasingly public platform to counteract all forms of racism. At least as
formidable as his contributions to anthropology was Boas's role in chal-
lenging the still-prevalent views in the United States that members of one
social or ethnic group were more intelligent or morally virtuous than
another. That there was no scientific basis for labeling one group as inferior
to another was the theme that Boas kept reinforcing in his writings. It is
a theme that became part of the fabric of social science from the 1930s until
the present.
Reactions to Boas
As Boas's contributions were primarily methodological, and as he had
an instinctive distrust of overarching theories, it is not surprising that his
most vocal critics in the next generation were those with a strong theoreti-
cal position to defend. Leslie White (1963) and Marvin Harris (1968),
devotees of evolutionism who were sympathetic to Marxism, portrayed
Boas as one who refused to take a stand on the relationship between one
culture and another, and who, in his passion for data about particular
individuals and groups, neglected the material and technological basis of
human activities. A. R. Radcliffe-Brown (1940, 1952), one of the leading
theorists on the British anthropological scene, stressed the importance in
anthropology of an undergirding theory; he promoted a Durkheimian
approach, in which the needs for group solidarity exert a decisive impact
on kinship structures and on the actions and beliefs of individuals. Rad-
cliffe-Brown also saw cultures as part of a social system, as "organisms"
which evolve toward increasing diversity and complexity. All of this theo-
rizing did not sit well with Boas, who distrusted generalizations about the
needs of a group or about a culture as a whole.
Another line of research was the functionalist approach of Bronislaw
Malinowski (1961, 1968): akin in some ways to behaviorism in psychology
233
and to Bloomfield's structural approach in linguistics, the Malinowski

approach evinced little interest in mental phenomena or in historical fac-
tors. Instead it focused almost exclusively on the painstaking description
of social and sexual behavior. According to Malinowski, the anthropolo-
gist should search for the various goals that a particular custom, material
object, idea, or belief serves within a society: the careful and imaginative
field work methods developed by Malinowski were devoted to this utili-
tarian end. Malinowski also went further than his predecessors in present-
ing a comprehensive view of native life-detailing the various practices
within a community and suggesting how they fit together. Malinowski's
striking ethnographies ended up having considerable influence on the
workaday practices of anthropologists, but his biologically and psycholog-
ically oriented explanations never captured the field: and Boas, himself a
careful fieldworker, wondered what all the fuss was about.
While Boas was not without his critics, most American anthropolo-
gists adopted his general program. Moreover, Boas was quite generous to
those who emerged from his shop. One student of his, Ruth Benedict
(1934), went much farther than Boas would have dared in laying out the
"character" of a civilization and in attempting to grasp the "meaning of
culture" as a whole. Another student, Alfred Kroeber (1917, 1948) stressed
the historical components of anthropology, pondered the concept of cul-
ture in its largest sense, and questioned the utility of focusing on particular
groups or particular behavior. Though Boas had doubt about Benedict's
ambitious program, and did not share Kroeber's various qualms about a
scientifically oriented anthropology, he continued to support the work of
his own proteges. In the struggle for the future of anthropology in Amer-
ica, personal loyalty was more potent than ideological consistency.
The Special Status of Language and Linguistics
While Kroeber had been selected by Boas to preside over anthropology in

the United States, Edward Sapir had been his chosen vehicle as the leader
of linguistics. Sapir shared Boas's interest in (and his gifts for) the record-
ing of Indian languages. He also agreed with Boas that all languages are
equally complex, and that (in their most fundamental respects) they de-
velop free of environmental determinants. Moreover, unlike many of
Boas's other students, Sapir had an enduring interest in the nature of
mentality in different cultures. But as a consequence of his own studies,
he came to conclusions at odds with Boas's sentiments. As I noted in
234
discussing the history of linguistics, Sapir and his student Benjamin Lee
Whorf came to believe that the language used by a group has been a
principal determinant of the belief structures and the ways of thinking of
that population. Indeed, Sapir came to see language as a guide to social
reality:
It powerfully conditions all our thinking about social problems and processes .
. . . It is quite an illusion to imagine that one adjusts to reality essentially without
the use of language and that language is merely an identical means of solving
specific problems of communication or reflection .... No two languages are ever
sufficiently similar to be considered as representing the same social reality. The
worlds in which different societies live are distinct worlds, not merely the same
worlds with different labels attached. (1929, p. 162)
The direction in which Sapir developed the relationship between lan-

guage and culture was fascinating, not truly anticipated in Boas's own
work but one that he came to admire as provocative. (Whether he would
have endorsed the more radical directions in which Whorf eventually
ventured is far less certain.) But there is another approach in anthropology,
also traceable to Boas's influence, where language as a system of analysis
came to be applied to cultural phenomena. This is the variety of structural
anthropology pioneered by Claude Levi-Strauss, one of the most eminent
anthropologists of our time-a scholar who had become an intimate of
Boas in the early 1940s, when the senior scholar's career was ending, and
the younger's was just on the rise.
Levi-Strauss has indicated his indebtedness to Boas's influence: "Boas
must be given credit for defining the unconscious nature of cultural
phenomena with admirable lucidity. By comparing cultural phenomena to
language ... he anticipated both the subsequent development of linguistic
theory and a future for anthropology whose rich promise we are just
beginning to perceive" (1963, p. 19). In his observation that neither lan-
guage nor culture rise to consciousness, Boas had discerned a vast opening
for anthropology. Precisely because people are unaware of these structures,
they are much less likely to revise them at will or to invent ad hoc explana-
tions of their nature and operation. But if Boas anticipated the general
direction in which anthropology should proceed, it was a mutual friend,
Roman Jakobson, who actually introduced Levi-Strauss to those insights
about language that would prove decisive in his own work (Jakobson 1941;
Jakobson and Halle 1956).
As a founding member of the Prague school of linguistics, Jakobson
had pointed out that certain properties of the human mind determine the
ways in which language operates. These properties may not be immedi-
ately evident; but, once specified, they allow the analyst to make sense of
235
II I THE CoGNITIVE SciENcEs : HISTORICAL PERSPECTivE
diverse linguistic phenomena. On Jakobson's account, because human be-

ings tend to perceive things in terms of polarities, many important distinc-
tions in language also prove to be binary. Phonemes are constituted of
distinctive features, with each distinctive feature either being present (for
example, voiced) or absent (for example, voiceless). Other aspects of lan-
guage, including grammar and meaning, can also be seen in terms of the
presence or the absence of various binary features. These features do not
exist in an unorganized fashion: they constitute a system, where the rela-
tions obtaining between the features become primary. Indeed, it is in
laying out the systematic and structured relationship among the various
terms that linguistics does its primary work.
The Structuralist Version
Levi-Strauss's Canons
This much was straight linguistic theory, of the sort put forth by
Jakobson and his associates in the Prague school and subscribed to, in more
or less faithful fashion, by others influenced by Saussure. The decisive step
taken by Levi-Strauss, and the core of his contribution to anthropology,
lay in his insistence-and his demonstration-that key aspects of culture
are best thought of as linguistic in nature and are best approached by the
methods of the structural linguist.
Appropriately for one who wanted to capture the attention of his
anthropological colleagues, Levi-Strauss began his inquiries by confront-
ing an area central to the concern of all anthropologists-the area of kin-
ship relations or kinship structures. To start with, he noted that in any
kinship system one has as primary data both the system of relations
between terms (father, son) and the system of relations between attitudes
(intimate, distant). As a test case for his notions of structural anthropology,
Levi-Strauss selected the classic problem of the avunculate: the relation-
ship where the maternal uncle represents family authority and exerts
certain rights over his nephew, and yet can maintain an informal, joking
relationship with that nephew. Levi-Strauss notes a correlation between
this set of attitudes and the young male's attitude toward his father. In
groups where familiarity characterizes the relationship between father and
son, uncle and nephew have a relationship of formal respect; whereas
when the father represents family authority, it is the uncle who is treated
with familiarity.
236
Radcliffe-Brown, a keen analyst of kinship structures, had himself
called attention to this phenomenon. But, said Levi-Strauss in response to
Radcliffe-Brown, it does not suffice to study the correlation of attitudes
between father and son and between uncle and sister's son. Rather, playing
structural linguist, one must take into account all the relevant terms and
the relationships among them all. In this particular case, the four crucial
relations are those between brother and sister, husband and wife, father
and son, and mother's brother and sister's son (Levi-Strauss 1963, p. 42).
Even as a linguist studies the phonological relations across many
languages in order to determine the proper set of distinctive features,
Levi-Strauss examined the avunculate in many cultures in an effort to
discover the operative factors. He then went on to propound a structural
law that, in his view, ferrets out the critical factors operating in this
complex set of relationships. The law reads: the relation between maternal
uncle and nephew (whether it be intimate or formal) is to the relation
between brother and sister as the relation between father and son is to that
between husband and wife. According to Levi-Strauss, if one knows one
pair of relations, it is possible, as in any analogy, to figure out the other
(1963, p. 42); he then illustrates his "law" by reviewing supporting exam-
ples. Levi-Strauss goes on to maintain that, even in quite different forms
of descent, one always encounters the same fundamental relationship be-
tween the four pairs of oppositions required to construct the kinship
system. One has thus unlocked the key to such relationships by having
figured out the appropriate unit of analysis. As he says, "This structure is
the most elementary form of kinship that can exist. It is, properly speaking,
the unit of kinship" (p. 46).
Levi-Strauss's early writings on kinship, while certainly controversial,
established his mark as a major anthropological thinker and also promoted
the injection of linguistic techniques (and, to a lesser extent, formal logical
analysis) into the research carried out by anthropologists. Acclaim also
greeted a second wave of work in social organization, another staple inter-
est of anthropologists. Here Levi-Strauss clarified the obscure nature of
dual organizations-where two parallel kinds of clans, often exogamous,
exist within the same village. Levi-Strauss adduced evidence that these
dual organizations actually mask the underlying dynamic force, which
arises from the exchange of women and other commodities. It is this
exchange, rather than the external residence patterns, that reflects the
actual social relations found in the village.
In the early 1950s, Levi-Strauss attended a conference of linguists and
anthropologists where he presented some of the material I have just out-
lined. During his remarks, Levi-Strauss alluded to an "uninvited guest
which has been seated during this conference beside us and which is the
237
human mind" (1963, p. 7). Levi-Strauss felt that in their focus on the material
and social organizational aspects of culture, anthropologists had given
short shrift to the key factor involved in understanding any culture-the
ways in which the human mind takes in, classifies, and interprets informa-
tion. Not coincidentally, Levi-Strauss was pointing out the need for an-
thropologists to consider mental representations just as other pio-
neering cognitivists were beginning to redirect efforts within their own
chosen disciplines-the Zeitgeist was assiduously at work.
Exploring Mind
Levi-Strauss's remark was prophetic concerning his own work, be-
cause for the remainder of his career he has sought to discover the nature
of the human mind in as pristine a form as possible. He has approached
this assignment by studying the ways in which individuals classify objects
and elements, and the ways in which they create and understand myths.
Much of this work is put forth as being empirical-based on the classifica-
tory systems observed around the world and on the myths related in many
Indian tribes. Yet, Levi-Strauss makes no secret of the fact that he must
rely on his own intuitions (which he has dubbed "neolithic"): introducing
his major study of myths he has even declared, "It is in the last resort
immaterial whether in this book the thought processes of the South
American Indians take shape through the medium of my thought, or
whether mine takes place through the medium of theirs" (1969, p. 13).
After all, he believes, it is the same mind-all human minds-in either
case, and the scholar's point of entry is simply not crucial. This may be
cognitive science-but it is a science built on a Cartesian confidence in
one's own mind rather than on the methods of consensual validation
embraced by nearly all other workers today.
While I cannot follow through all the steps of Levi-Strauss's complex
and still uncompleted project, I can say something about the methods he
uses and the conclusions he reaches. In his studies of classification, he
comes down decisively in favor of the proposition that the principal fea-
ture of all minds is to classify, and that primitive individuals classify pretty
much along the same lines, and in the same ways, as the most civilized
persons. He describes the classifying practices of primitive groups as a
science of the concrete: rather than looking for the factors that underlie the
structures or processes of the world (as the trained scientist does), the
primitive mind seeks to classify everyday objects and experiences in terms
of their overt perceptual and sensory properties. These methods do not
always lead to the same categories and classes as those used in the Western
scientific approach-they may be more or less detailed and may have
238
differently drawn boundaries; but they reflect roughly the same kinds of
analytic moves on the part of a classifier.
Nor is there an unlimited number of ways in which the human mind
can work. As humans, we are constrained in the kinds of combinations we
can make, in the kinds of distinctive features of opposition with which
we can play. Levi-Strauss laid his cards on the table in this procla-
mation:
The ensemble of a people's customs has always its particular style; they form
into systems. I am convinced that the number of these systems is not unlimited and
that human societies, like individual human beings (at play, in their dreams, or in
moments of delirium), never create absolutely: all they can do is to choose certain
combinations from a repertory of ideas which it should be possible to reconstitute.
(1964, p. 60)
Though the degree of analytic precision diverges from that encountered in

contemporary linguistics, it is germane to compare Levi-Strauss's Men-
delian notions here with the limits on information processing proposed by
George Miller in the middle 1950s or the "setting of parameters" now
favored in Chomskian linguistics.
Myth Making
Levi-Strauss's studies of myth making represent his most extensive
search for the rules governing human cognition. In an early work (1963),
he laid out a methodological approach for the structural study of myth.
Proceeding in approved structuralist fashion, he proposed a breakdown of
a myth into component parts or units (the elementary phrases of the myth)
and then the assembling of all units that refer to the same theme or make
the same point. For example, in the case of the Oedipal myths, Levi-
Strauss discerns a set of themes relating to the overvaluing of blood rela-
tions (Cadmos seeks his sister, who has been ravished by Zeus; Oedipus
marries his mother, Jocasta); a set of themes relating to the undervaluing of
blood relations (Oedipus kills his father, Laius; Eteocles kills his brother
Polynices); a set of themes relating to monsters being slain (Cadmos kills the
dragon; Oedipus kills the sphinx); and some unusual names having to do with
difficulties in walking (Labdacos means "lame"; Laius means "left-sided";
Oedipus means "swollen-footed").
Having grouped the various myth themes into these four categories,
"Levi-Strauss then lays out a formula that purports to describe the underly-
ing message of the myth. As he describes it, the Oedipus myth in all its
myriad versions has to do with either the overvaluing or the undervaluing
of the importance of kinship structure, and with the question of men's
239
I I I THE CoGNITIVE SciENCES : HISTORICAL PERSPECTivE
origins on the earth either through autochthony (emergence from the

earth, with certain creatures having to be killed so that men can be born
from the earth) or through childbirth (men from the earth cannot walk
initially or they walk clumsily). The myth does not resolve these issues-
where men actually come from-inasmuch as myths deal with perennial
mysteries; yet it does provide a point of equilibrium by laying out various
competing themes and suggesting some balance among them. Ultimately
Levi-Strauss describes the Oedipal myth in this formulaic way: the over-
rating of blood relations is to the underrating of blood relations as the
attempt to escape autochthony is to the impossibility of succeeding in it
(Levi-Strauss 1963, p. 216).
While this elliptical example may seem forbidding to follow, and even
more difficult to swallow, it actually conveys rather well the kind of
enterprise in which Levi-Strauss has been involved. To attempt to "trans-
late" his account of a myth structure, or a kin structure like the avunculate,
so that it is clearer and less exotic than the original is as distorting a
maneuver as would be a deliberate obfuscation of his argument. As one
can see, he is willing to take any kind of myth fragment or myth corpus
and reduce and rearrange its elements, in order to come up with an account
of the themes with which it is working and the kinds of messages that seem
implicit in it.
But what is Levi-Strauss up to? As I see it, he believes that the simple
empirical categories that populate myths-percepts of smell, sound, si-
lence, light, darkness, rawness or cookedness-are best conceived of as the
conceptual tools for approaching the more abstract concerns with which
human beings everywhere must grapple: dilemmas like the dialectic be-
tween nature and culture, the status of the incest taboo, the relation be-
tween sexes, the importance of particular social arrangements. These ideas
are stated in terms of concrete particulars, such as emotionally laden stories
of social conflicts, but they can be reformulated at a more abstract level in
terms of logical propositions. In fact, if the myths are to be properly
understood, the logical terms and relations must be specified, as Levi-
Strauss has attempted to do in his gloss of the Oedipal myth. Moreover,
he goes on to insist that the myths themselves have a quasi-biological
relationship to one another. Each myth in a sense transforms the others
that go before it, and no myth can be fully understood except in relation
to all the other myths in a corpus (which might ultimately include all the
myths ever spun). The kinds of relationship among phonemic features that
define a linguistic system are mirrored in the thematic forms that constitute
a mythic system.
While myths may confuse many of us, they offer a privileged route
for a Uvi-Straussian analysis. As he sees it, the myth offers a glimpse of
t~ mind in pure form: "when the mind is left to commune with itself and
240
no longer has to come to terms with objects, it is in a sense reduced to
imitating itself as object" (1969, p. 10). And his examination has confirmed
for him the essential logic inherent in all human thought:
The kind of logic in mythical thought is as rigorous as that of modem science,

and ... the difference lies, not in the quality of the intellectual process, but in the
nature of things to which it is applied .... Man has always been thinking equally
well; the improvement lies, not in an alleged progress of man's mind, but in the
discovery of new areas to which it may apply its unchanged and unchanging
powers. (1963, p. 230)
And so Boas's program is vindicated by Levi-Strauss's myth analysis.

It should come as no surprise that Levi-Strauss's work, while it has
dazzled most readers, has not convinced many of them. It is too neat
somehow that all of the myths of the world, which could have arisen from
such diverse sources and for such varied reasons, should converge to em-
body the various messages that one thoughtful but solitary Frenchman
discerns in them. Levi-Strauss himself varies in the actual claims he makes
for his analyses, as in the preceding quotation. But, over all, it is his feeling
that while kinship is too embedded in social action to provide a sure guide
to mental processes, mythology holds the key to unlocking the laws of the
mind. That systems can be found is for Levi-Strauss the evidence that the
mind has its laws of order that the anthropologist-or, he might as well
say, the cognitive scientist-is challenged to discover.
Like Noam Chomsky, Levi-Strauss has stimulated a great deal of
admiration for the scope of his project and the daring way in which he has
sought to reorient his field. Unlike Chomsky, his method cannot be laid
out with sufficient clarity so as to be followed by anyone who exhibits
good faith and hard work. Thus, rather than having a school of followers,
Levi-Strauss has chiefly imitators, who seek to carry out the kind of intui-
tive subjective analysis in which the master excels. Of these, the English
anthropologist Edmund Leach is probably the best known as well as the
most gifted (1961, 1974). But Levi-Strauss shares with many original
thinkers the desire either for complete fidelity (on the part of others) to his
own system or for some formal declaration of distance. Thus, many of
those initially sympathetic to him, including Leach, have eventually been
left out in the cold-a situation not, apparently, displeasing to Levi-
Strauss.
Is Levi-Strauss, then, a pivotal contributor to cognitive science or just
an isolated, humanistically oriented savant? Most of my colleagues would,
I suspect, consign him to the tradition of Rousseau, Montesquieu, and the
French Academy (to which he was recently elected) rather than to the
ranks of Simon or Chomsky, but I think that this assessment will turn out
to be shortsighted. In my own view, despite his conceits and idiosyncra-
241
I I I THE CoGNITivE SciENcEs : HISTORICAL PERSPECTIVE
cies, his rather cranky set of interests and obsessions, Levi-Strauss will
prove to be an enduring figure in the history of cognitive science. At
mid-century he injected issues of cognition centrally into anthropological
discussions. By invoking the most rigorous approaches of linguistics dur-
ing his time, and applying them imaginatively in the principal domains
within anthropology, he opened up new fields of inquiry. Like Piaget, he
has sometimes aped system, or haphazardly borrowed terms from formal
analysis, rather than being truly systematic: it is therefore up to his succes-
sors to pursue his program in a less personal and more replicable way
(though how to do this is much less clear than, say, in the case of Piaget).
Again, as happened with Piaget, Levi-Strauss's strong positions on an-
thropological issues have served as a convenient point of departure for
revisionist formulations.
When all the criticisms have been considered, few of Levi-Strauss's
particular conclusions may remain; but the critiques will only have been
possible because of his fertile questions and provocative hypotheses.
Moreover, I suspect that one hundred years from now his research program
will be seen as more right-headed than that of his strongest critics-the
true mark of an important thinker. Levi-Strauss will endure because he
posed questions that are central to both anthropology and cognition; out-
lined methods of analysis that might be applicable; and proposed the kinds
of systematic relations which may obtain in such diverse fields as kinship,
social organization, classification, and mythology.
Sperber s Variations
Of those who have some sympathy with Levi-Strauss's enterprise,
Dan Sperber, a former student of his now working in Paris (1968, 1982),
endorses his teacher's once radical program of examining the products of
human mentation and concurs that the model of linguistics is crucial for
anthropologists. Yet, perhaps appropriately for one of a younger genera-
tion, Sperber feels that Levi-Strauss has drawn on the wrong school of
linguistics. The structural approach is as outmoded in anthropology as it
has become in linguistics, maintains Sperber; and it is from the work of
Chomsky, Fodor, and others of the transformationalist school that the
anthropologist must now seek models.
Here the lessons tum out to be largely negative, for language is seen
as a system that works by its own very special rules. Contrary to Levi-
Strauss's vision (and reflecting the shift toward domain-specific princi-
ples), Sperber maintains that linguistic analysis cannot properly be applied
to other cultural phenomena such as myths, custom, and rituals. One
should instead consider cultural phenomena as entities subject to an end-
less amount of mental associations or elaborations, of the sort which go on
242
precisely a/fer the usual rule-governed operations of language have been
carried out. It is in the realm of the mysterious, the unanalyzable, the richly
symbolic, that most cultural entities exert their powers-and not in the
relatively lean aspects of language which can be apprehended and analyzed
according to prescribed syntactic, phonological, and lexical regularities.
Sperber points out that most human beliefs are not purely propositional
but are rather semipropositional-not fully logical and far fuzzier. It is
risky to apply to such amorphous belief systems the rigid classificatory grid
of the syntactician or the phonologist. Instead, one needs to study the
processes whereby rich penumbras of meaning are evoked.
Sperber's positive contribution inheres in his characterization of sym-
bolic processes. Rather than being induced or constructed from experience,
the symbolic mechanisms are part of the innate mental equipment which
makes experience possible. These mechanisms of symbolic elaboration,
working in a manner reminiscent of Levi-Strauss's savage, start with the
assumption that the "waste" of the mind ought always to be salvaged
because something can be made of it. It is just because this dross harbors
within it atypical or exceptional conceptual characteristics that it lends
itself to unending symbolic elaboration. In such cases, the symbolic mech-
anism of the mind draws on one's encyclopedic knowledge, one's knowl-
edge of more or less remote categories, and, indeed, any other modes of
information or interpretation that happen to be available-all in an effort
somehow to piece together these disparate elements into an overall sensi-
ble framework. Anthropology is the discipline that has access to the fullest
range of beliefs, practices, and symbolic systems; hence, it is in a privileged
position to lay bare the operation of those human symbolic mechanisms
that supplement the (relatively) pure computational aspects involved in
language, mathematics, and ordinary classification.
There is no shortage of critics of Levi-Strauss, who fault him at every

step, from the initial ethnographies he published in the 1940s to the dubi-
ous importation of methods from cybernetics, information theory, or lin-
guistics which crowd his pages. To my mind, the most telling line of
criticism questions whether polymorphous human behavioral patterns and
beliefs can lend themselves to the kind of systematic, rule-bound, and
"closed" analysis that has proved appropriate for certain aspects of linguis-
tic structures. In the view of anthropologist Clifford Geertz, Levi-Strauss
has jumped prematurely to the conclusion that human thought processes
can be analogized to the operations of a traditional computer (Geertz 1973,
1983). The preferred path, in Geertz's view, lies in the careful detailed
study of a social group in its cultural setting: only in this way can one hope
to understand the kind of symbolism it uses and the way in which it goes
about making sense of the world. Geertz criticizes Levi-Strauss's mechan-
243
II I THE CoGNITIVE SciENcEs : HisToRICAL PERSPECTIVE
istic approach, his ignorance of the particular historical conditions that

spawn a given myth or social organization, the minimization of affective
and emotional factors, the loss of the specific individual with his or her
own needs, motivations, goals, and wishes-all in favor of an intellect that
grinds out classifications and propositions. Geertz also questions the wis-
dom of construing symbolic products as the output of internal cognitive
mechanisms: according to his more public view of mind, myths, rituals,
beliefs, and the like are a socially generated form of symbolization:
The way to look at thought is not to assume that there is a parallel thread of
correlated affects or internal experiences that go with it in some regular way. It's
not of course that people don 'f have internal experiences, of course they do; but that
when you ask what is the state of mind of someone, say while he or she is
performing a ritual, it's hard to believe that such experiences are the same for all
people involved .... The thinking, and indeed the feeling in an odd sort of way,
is really going on in public. They are really saying what they're saying, doing what
they're doing, meaning what they're meaning .
. . . Thought is, in great part anyway, a public activity. (Quoted in Miller 1983,
pp. 202-3)
Whether the program of structural anthropology might ultimately make

sense is not directly addressed; but it seems clear to critics like Geertz that
the time for such an assault on the universals of thought is premature. We
must first understand individual cases much better than we do; and that
understanding must encompass ourselves no less than the "funny" people
elsewhere whose difficult-to-pigeonhole practices have perennially stimu-
lated the symbolic processes of anthropologists.
Ethnoscience
What Levi-Strauss undertook pretty much by himself, armed with his own
intuitions, has in the last three decades become in somewhat transmuted
form a major wing of anthropology. I refer here to the field of ethnoscience,
in its various lexical guises-componential analysis, ethnosemantics, cog-
nitive anthropology-all comprising the organized study of the thought
systems of individuals in other cultures and sometimes in our own.
Roots
The factors that gave rise to ethnoscience in the United States in the
middle 1950s resembled those that stimulated Levi-Strauss in France at
about· the same time. Both anthropological circles were affected by the
244
example of linguistics as a social science that had achieved unparalleled
rigor. Information theory, stochastic modeling, cybernetics, and computer
simulation were other reminders of the advantages to be gained by formal
approaches to social phenomena. Thanks to the works of the semioticians
like Roman Jakobson (1963} and Charles Morris (1971}, there was increas-
ing awareness of the essentially symbolic property of all human action, and
of the need to explain it in terms of its cognitive, as opposed to its practical,
aspects. According to one of the first ethnoscientific practitioners:
The model for the first experiments in the structural analysis of meaning was
consciously an analogical adaptation of that which had been developed for pho-
nemic analysis. More than one of those who were involved in the early phases of
this experiment have acknowledged that the stimulus to it, as well as the eye-
opener as to how it might be done, came from a combination of their training in
phonemics and their reading of Charles W. Morris' Foundation of the Theory of
Signs. (Lounsbury 1968, p. 223)
In fact, this background spawned the initial pair of publications con-

sciously styled in the ethnoscientific mode: Ward Goodenough's (1964)
revision of his analysis of the Trukese kinship terminology, which he had
undertaken in the field some years earlier (1951, 1956); and Floyd Louns-
bury's (1956) structural analysis of the Pawnee terminology collected for
Lewis Henry Morgan by an Indian agent in 1863.
Just how have the practitioners of this new form of science conceptu-
alized themselves? According to Stephen Tyler, the first anthologist of
cognitive anthropology, the field focuses on discovering how different
peoples organize and use their cultures.
It is an attempt to understand the organizing principles underlying their

behavior. It is assumed that each people has a unique system for perceiving and
organizing material phenomena-things, events, behaviors, and emotions.... The
object of the study is not these material phenomena themselves, but the way they
are organized in the minds of men. Cultures then are not material phenomena: they
are cognitive organizations of material phenomena. (Tyler 1969, p. 3)
Drawing on the model of a linguist's grammar, ethnoscientists search for

the ways in which knowledge of a culture's rules is reflected in the behav-
ior of natives, and especially in their speech.
Anthony Wallace, an early worker in the field, compares the eth-
nosemanticist to an observer with the following assignment: describe a
calculus being used by a group of people who have not formulated their
system in a written text. The investigator is allowed to interview and
observe, and to learn the method as a novice himself. As he analyzes that
data, he does not merely tabulate frequencies and give equal weight to
every slip, joke, gibberish, or error:
245
What he does, instead, is to infer the system of rules that these people are
attempting to apply. He will gain the assurance that he is on the way to an adequate
understanding of these rules from the logical completeness of the system he infers
and from his ability, when using it, to produce behavior that an expert will reward
by saying, in effect, "That's right; that's good; now you've got it." (1968, p. 537)
Robert B. Edgerton and L. L. Langness have put it even more succinctly:

"The goal of an ethnoscientific description is to write a set of rules for a
culture which is so complete that any outsider could use them to behave
appropriately in that culture" (1974, p. 38).
A Sample Componential Analysis

A common ethnoscientific approach is componential analysis. Toil-
lustrate how a componential analysis might actually be carried out, I
have borrowed an example from an early study of English kin terms by
Anthony Wallace and John Atkins (1960). These authors begin by taking
a set of kinship terms: namely grandfather, grandmother, father, mother,
brother, sister, son, daughter, grandson, granddaughter, uncle, aunt,
cousin, nephew, and niece. Next, they define these terms with respect to
genealogical relations. Thus, grandfather is defined as someone who is
one's father's father (or, in the notation used by anthropologists, Grand-
father: FaFa). The expression "Grandfather: FaFa, MoFa" is translated as
"grandfather refers to 'father's father' and to 'mother's father.'" All
terms are defined through the primitive forms Fa, Mo, Br, Si, So, Da, and
some simple operators.
The rest of a kinship grid contains entries like:
Grandmother: FaMo, MoMo

Grandson: SoSo, DaSo
Uncle: FaBr, MoBr, FaFaBr, MoFaBr, and so on,
culminating in the entry after cousin which includes at least twelve possible
relatives. In this particular kinship grid, the terms uncle, aunt, nephew,
cousin, and niece have been used in an extended sense, thus including such
relations as "second cousin once removed" and other distant relations
within an extended family.
The third stage entails a number of observations obtained from the
grid. For instance, all but one of these terms (cousin) specifies the sex of the
relative: some (like grandfather but not cousin) specify generation; all specify
whether the relative is lineally (for example, sons) or nonlineally (for exam-
ple, nephew) related to Ego; and nonlineal terms specify whether all the
246
ancestors of the relative are direct ancestors of Ego, whether all the ances-
tors of Ego are ancestors of the relative, or neither.
Now comes the crucial stage: the analyst hypothesizes that three
dimensions will be sufficient to define all the terms. The sex of the rela-
tive (male as a 1, female as a 2 }; generation (b 1 as two generations above Ego;
b2 as one generation above Ego; b3 as Ego's own generation; b4 as one
generation below Ego; b5 as two generations below Ego); finally, linea/-
ify: c1 is lineal; c2 is co-lineal (brother or sister); c3 as ablineal (cousin).
Lineals are persons who are ancestors or descendants of Ego; co-lineals are
nonlineals, all of whose ancestors include, or are included in, all the ances-
tors of Ego; ablineals are relatives who are neither lineals nor co-lineals.
In the next step the terms are now redefined as components. Thus, the
grandfather is a 1 b 1 c1; the grandson is a 1 b5 c1; the sisteris a 2 b3 c2; an uncle
is a 1 b1 c 2 and a 1 b2 c2 ; and a cousin is a b (not marked on either sex or
generation) and c3 . (Note the convention that, when a term does not
discriminate on a dimension, the letter for that dimension is given sans
subscript).
It is now possible to summarize this analysis of kin terms by means
of the paradigm in the following table. Conforming to the technical defini-
tion of a paradigm, each term has been so defined that no term overlaps
or includes another; every component is discriminated by at least one term;
and all terms can be displayed on the same paradigm.
ci Cz CJ
al az aI az aI az
grandfather grandmother
uncle aunt
father mother
[ego] brother sister cousin
son daughter
nephew niece
grandson granddaughter
Source: From A. F. C. Wallace and J. Atkins, "The Meaning of Kinship Terms." Reproduced
by permission of the American Anthropological Association from American Anlhropologisl 62
(1): 58-80, 1960.
247
This analysis accomplishes the goal of laying out the relationships

among the various kin terms in our language in a clear, exhaustive, and
simplifying way. Wallace and Atkins make no claims that this is the best
way, or the only way, to lay out the components of American kin structure
(for an influential rival account, see Romney and D'Andrade 1964)-only
that it is one possible and parsimonious way of accomplishing this task.
Wallace and Atkins's examples also convey the concerns that occu-
pied the minds of the founders of componential analysis. These researchers
felt that it made sense first to apply the method in an area that is already
relatively well understood; where there is a finite number of terms that
could be defined on independent grounds (like sex); and where societies
differ in potentially revealing ways from one another in the kinds of
terminology they use, and the ways in which these societies cut up the
kinship pie. It was also important that members of those societies hold
shared intuitions about their respective kinship structures. For example,
among the Iroquois the primary dimensions are those of generation and
bifurcation (a distinction between parallel and cross-cousin relations);
whereas in Crow kin systems, the primary dimensions are a skewed generation
measure (where females rank a social generation higher than their male
siblings in certain contexts) and bifurcation (Lounsbury 1963, p. 572). Floyd
Lounsbury declared:
We may consider that a "formal account" of a collection of empirical data has

been given when there have been specified 1) a set of primitive elements, and 2)
a set of rules for operating on these, such that by the application of the latter to
the former, the elements of a "model" are generated; which model in tum comes
satisfactorily close to being a facsimile or exact replica of the empirical data whose
interrelatedness and systemic nature we are trying to understand. A formal account
is thus an apparatus for predicting back the data at hand, thereby making them
"understandable," i.e. showing them to be lawful and expectable consequences of
an underlying principle that may be presumed to be at work at their source.
(Quoted in Durbin 1966, p. 31)
Goodenough adds a less formal note:
One test of the adequacy of this account, I have said, is that it does not do
violence to my own feel, as informant, for the structure of what is described. This
is the subjective test of adequacy. An equally important test is that it provide an
alien with the knowledge he needs to use my kinship terminology in a way I will
accept as corresponding with the way I use it. This is the objective test of adequacy.
(Quoted in Kuper 1973, p. 573)
Thus described, componential analysis may seem a relatively simple

and straightforward process on whose application trained anthropologists
248
should agree and by which they should ultimately arrive at the same
conclusion. And indeed, at least compared with Levi-Strauss's avowedly
subjective measures, it is possible to train students in componential analy-
sis. In fact, however, componential analysis turns out to be more complex
and "fuzzier" than its originators had hoped.
Critiques of Ethnoscience: From Within

The area of kinship represents a positive extreme with respect to the
feasibility of listing terms and the likelihood of figuring out the relevant
dimensions in a reliable fashion. When one turns to other domains, ranging
from color, to botany, to disease, it turns out to be more complex to elicit
the relevant terms and delineate the domain, let alone to ferret out the
relevant dimensions that may systematize the domain in a defensible and
desirable way.
Even when the terms and dimensions have been delineated, the way
in which to arrange them becomes a subject of considerable controversy.
The clear-cut example of a paradigm I have just illustrated is but one of
the numerous formal ways in which terms can be arrayed. There are also
diagrams, mazeways, taxonomies, trees, etic grids, contrast sets, and many
other ways of laying out the findings: the relationship among these ways
of delineating is by no means clear, and it is possible to make combinations
of them or to lay out findings in more than one way, as in box or key
diagrams. Even English kinship terms can be laid out in ways other than
paradigms, depending on the terms used, the methods used to elicit the
definitions, and the ways in which the dimensions are aligned with refer-
ence to one another (Kay 1966).
Still thornier questions arise when one wants to determine whether
a componential analysis is appropriate, or which of a number of competing
analyses is most accurate. The intuitions of a native speaker or of a judi-
cious anthropologist provide one measure, but it is also possible to use
empirical methods-for example, requiring informants to answer ques-
tions, to sort items, to define terms, to group words, or even to construct
their own componential analyses. The "psychological reality" of an analy-
sis-is it in the heads of all informants, of trained and reflective infor-
mants, or only in the head of the analyst?-turns out to be one of the most
complex questions in componential analysis. Indeed, one can raise, with
respect to a "simulation" of a native's knowledge, the same questions
posed by John Searle in his "computer-Chinese room" conundrum.
Some of these critical points were brought up early on by Wallace and
Atkins (1960), who pointed out problems with homonyms and metaphors,
where the same words might have different meanings, or where different
249
I I I THE CoGNITIVE SciENCEs : HISTORICAL PERSPECTivE
words might have the same meanings. There was the problem of connota-
tion, where words may have the same objective meaning but connote
different affective values (dad, daddy, old man, pop, father, and so on). Other
critics have argued that componential analysis is inherently circular, since
one must begin by assuming the very relationship among terms whose
relationship should actually be fixed only at the conclusion of the investi-
gation. In other words, the existence and coherence of the domain is
presupposed before the investigation begins (Gardin 1965). Some com-
mentators have focused on the enormous problems of translating terms
from a foreign language (particularly one of non Indo-European origins)
into a familiar tongue and assuming that the same kinds of analysis can
be applied to the translations (Crick 1976; Keesing 1976). For example, in
a culture where men and women do few things together, it is risky to
transfer the same meanings to the foreign terms as we would apply to male
and female; and the terms male and female themselves have very different
frequency and usage than do man and woman (Lyons 1968).
Critiques of Efhnoscience: Outside the Ranks

Criticisms from those who are not aligned with formal analytical
approaches are more severe still. Clifford Geertz points out that knowing
how to talk about something is not the same as doing something. One has
to pay attention to the logic exhibited in actual life, not to some kind of
abstracted set of symbolic elements whose validity for the inhabitants--
let alone the formal logical relations that may obtain among them-is
questionable at best. "Nothing has done more, I think, to discredit cultural
analysis than the construction of impeccable descriptions of formal order
in whose actual existence nobody can quite believe" (Geertz 1973, p. 18).
Gary Witherspoon argues that many aspects of importance are simply not
marked in the language and concludes:
How a Navajo thinks and what categories he can employ in his thinking are
not the same.... The assumptions that kin terms have primary ... referents in
all cultures, that kinship and kinsmen are basically the same in all cultures, and
only partitioned differently by various sets of kin terms, and that kin terms corre-
spond to and express the kin categories of a given culture seem extremely naive.
(1971, p. 116)
Michael Silverstein (1978) views cognitive anthropology as a natural out-

growth of the importation of Bloomfieldian views into anthropology.
Words are seen as standing directly for things. Reflected is an utterly
atomistic conception of language, functionally ordered into data arrays,
without any sense of how words function in a social context, the kinds of
actions in which they are embedded, and the ways in which they interact
250
with and influence one another. Language as a structured system, not
language as a set of isolated tokens referring to isolated objects, should be
the model of choice for anthropologists.
Perhaps most telling in any discussion about the fate of ethnoscience
is the hegira of Stephen Tyler himself. Initially a staunch proponent of
componential analysis and its first anthologist, he became completely disil-
lusioned in the intervening decade. In 1978 he published The Said and the
Unsaid, in which he spurned the earlier view of language as offering a
reliable window on the cognitive systems of individuals. In his formulation:
Common to both logical positivism and transformational linguistics is their

view of language-as-mathematics. Both focus on language as a system of primitive
or elementary units which can be combined according to fixed rules. However
useful this analogy may be in certain limited ways, it creates problems in under-
standing how the purely formal system of elements and rules relates to something
other than itself. Both create dualistic systems which oppose formal linguistic
competence to empirical components. (1978, pp. 13-14)
Tyler now stresses the functionalist facets of language. On this account,

language is not merely a means of representing ideas but equally a means
for expressing wishes, feelings, and emotions-and, above all, a way of
getting things done in the world. As he phrases it, language is a means of
establishing relations rather than an object consisting of relations. What
we take to be someone' s intentions, purposes, plans, and attitudes are dues
to what that person means.
Tyler ends up endorsing a hermeneutic approach. In this perspective,
one looks at what is said and unsaid-the background of unspoken pre-
suppositions and implications-created conventionally by the said and
created intentionally by the speaker and hearer. He rejects the assumed
independence of semantics from pragmatics:
No less than the death of meaning should we have forecast from a manner
of thought that emptied thought of all content, and what else could we expect from
a method of analysis that presumed to show that meaning might mysteriously
emerge from the mechanical concatenation of meaningless elements? ... Whether
in art or science nothing is clearer than the intellectual poverty of formalism. (1978,
p. 465)
Tyler's benediction over componential analysis has not by any means

spelled the death of this movement. However, there is now abroad consid-
erable awareness that the original methods of componential analysis, while
suitable in some domains, may not be widely extendable. Such techniques
may be useful in studies of kinship and color; they may be plausible with
other possibly finite categories like plants, animals, or everyday objects;
but they rapidly become dysfunctional when applied to slippery areas, like
251
I I I THE CoGNITIVE SciENCEs : HisTORICAL PERSPECTIVE
emotions or diseases, where the line around the domain is not announced
in advance and where an individual's (or a group's) idiosyncratic interpre-
tive system comes more readily to the fore.
In the case of illness, for example, one's responses are so intertwined
with the values of one's culture, the nature of pain, the prevailing theories
of causation and cure of disease, the distinction between observable, and
experienced symptoms that the analyst rapidly loses confidence that di-
verse informants are talking about the same thing or even the same do-
main. Sensing these difficulties, Roger Keesing, another early supporter of
the new ethnography, declared as early as 1976:
For almost fifteen years, cognitive anthropologists have pursued "the new
ethnography" as far as it would lead them. For the last five, at least, it has been
obvious that this would not be very far-that the messianic promises of the early
polemic were not to be realized. "The new ethnographers" have been unable to
move beyond the analysis of artificially simplified and delineated (and usually
trivial) semantic domains and this has discouraged many of the originally faithful.
(1976, p. 307)
It is not by any means the first occasion in cognitive science where formal
methods turn out to work most effectively with those aspects of behavior
that seem least central to mankind's concerns. Clearly, the computational
paradox resonates in anthropological circles.
Stephen Murray (1982), a historian who has reviewed the rise and
subsequent decline of "classical ethnoscience," discerns two different rea-
sons for the evanescence of this form of study. The first stems from a
much-touted promise that was not fulfilled. In the middle 1960s, a major
study of drinking in Chiapas, Mexico, was designed to test the powers of
ethnoscientific research. According to the plan, investigators were sup-
posed to apply the same procedures for analyzing patterns of drinking to
data secured from five villages. However, problems of analyzing the data
were never solved and the investigators themselves eventually moved on
to other problems. As Paul Kay, one member of the team, reports, ·
It turned out that after collecting a huge amount of material and spending two
or three years looking for a set of objective procedures-or even semi-objective
procedures-that could be applied to this material to reduce it to some kind of
logical statement, we gave up, because we could not find such a set of procedures.
Drinking is an institution in Chiapas that permeates the entire lives of a people:
religion, politics, family life, even agriculture is inextricably tied up with drinking,
so to do the ethnography of drinking there is to do the total ethnography. (Quoted
in S. 0. Murray 1982, p. 169)
Kay's colleague Brent Berlin suggests, "We were not convinced that what
could be said from the elicited data was that much more revealing than
252
what could be said on the basis of old-fashioned participant observation"
(p. 169).
The experience in Chiapas may have signaled an important lesson for
anthropologists. While certain domains may permit themselves to be
bracketed from the rest of the culture, most domains of interest may be
inextricably bound up with the surrounding context. To study drinking
without studying everything else may (from the anthropological point of
view) be a scientifically untenable posture: in this case, to isolate is to
destroy. Berlin and Kay, who began as enthusiasts of the enthnosemantic
method, are here underlining its limitations. Contextual, historical, and
cultural effects may be of the essence in most of the anthropological
terrain.
Murray cites another reason for the decline of ethnoscience: that is,
this anthropological subspecialty never cohered into a single integrated
perspective but was, at best, a loose confederation. Murray believes that
self-styled revolutionary groups, which demand total commitment and are
organized under a single visionary leader, are more likely to prevail in
academic competition. (The example of Chomsky in linguistics springs to
mind.) A loose confederation offers less resistance to competitors and
ultimately ceases to exist as a recognizable approach. Instead, "having
achieved a measure of recognition and success, ethnoscience grouping
encountered the seeming price of success: splintering.... By the late 1960s
new students were not being attracted and classical ethnoscience was no
more" (1982, p. 172).
There is scarcely any consensus within anthropology about which
steps ought to be taken in lieu of a strict ethnoscience. Some authorities
have continued to carry out studies in the tradition but have substituted
more modest goals. Others have adopted the perspective of Clifford Geertz
(1973, 1983), who feels that anthropologists have erred in attempting to
mimic the natural sciences, and that the anthropologist has greater affinity
with an interpreting literary critic. (I shall consider a Geertzian critique
again in later discussions of cognitive science.) Still others feel that notions
and concepts developed in other areas of cognitive science can be usefully
imported into anthropological study.
Psychological Forays
While componential analysis made the biggest splash on the anthropologi-

cal scene, the importation of methods from other disciplines has also
flourished in past decades. In the tradition of the Torres Strait expedition,
253
some investigators have analyzed the perceptual capacities of diverse

populations. In an ambitious study, Marshall Segall, Donald Campbell,
and Melville Herskovits (1966) attempted to resolve a dispute about envi-
ronmental contributions to visual perception by studying the susceptibil-
ity of several groups to various optical illusions. They subjected individu-
als from fourteen non-European cultures to a variety of illusions.
Supporting the position that one's experiences influence the way one sees,
these investigators found that on certain items, such as the Miiller-Lyer
illusion,.. European and American samples proved more susceptible to
illusions. These seemed to reflect life in a "carpentered" environment,
replete with many square and rectangular objects. In contrast, on certain
other perceptual arrays, non-Western samples proved more susceptible to
the illusion. For example, among individuals who live in an equatorial
terrain, where one must look at flat and wide spaces, there is a greater
tendency to fall prey to the horizontal-vertical illusion. Thus, in an issue
long of interest to culturally sensitive students of psychology, these inves-
tigators were able to discern intriguing regularities.
A second line of psychologically inspired studies addresses the capaci-
ties of individuals from non-Western societies to reason in the manner of
a Westerner. In this regard the measures used by intelligence testers, and
also those developed by Jean Piaget in his work with children, have proved
most transportable. An early wave of studies was sympathetic to the
notion that people outside the West, and particularly those who had not
attended school, performed far more poorly on tests of abstraction, con-
ceptualization, and classification (Dasen 1972; Dasen and Heron 1981).
(Results were much more comparable on tests of perceptual and motor
skill.) But before it was possible to conclude that the non-Westerners were
less intelligent or could not think as well, some important methodological
adjustments were made by Michael Cole and his colleagues (Cole and
Scribner 1974). It emerged that, when familiar materials were used, or
when requested behavior was explained to or modeled for the subjects,
many of the documented differences between individuals from the two
cultures now evaporated.
As a result of these instructive demonstrations, much greater stress is
now placed on using materials that are "culture-fair" and on testing those
capacities that are basic, rather than those likely to be transmitted only in
schooled settings. Most evidence now suggests that, as Boas long believed,
the fundamental operations of thought are the same everywhere, and it is
the uses to which these processes are put that differ dramatically across
cultures. Superimposed upon this basic continuity is the advent of certain
•In the Miiller-Lyer illusion, one of the two lines >~<t::~)~< appears longer than
the other, even though both are exactly the same length.
254
abilities to reason without the usual contextual supports, or to carry
through certain complex chains of reasoning, which seem to develop
chiefly among individuals exposed to years of Western-style secular
schooling.
Investigators have been revisiting some of the issues about language
and thought that had been raised many years ago. While strong experi-
mental support for the Whorf-Sapir hypothesis had never been obtained,
there was a widespread suspicion among anthropologists that the differ-
ences between conceptual systems in remote cultures were vast and that
these might well reflect variations in the structure or the contents of
language. However, in a line of study which continues to exert wide
influence in several cognitive sciences, Eleanor Rosch (1973a, 1973h-then
Eleanor Heider) strongly challenged the Whorfian line: she demonstrated
that, even in cultures with few color terms, individuals still sort, classify,
and otherwise deal with the color spectrum in roughly similar ways. The
language does not affect basic psychological processes. Paralleling this line
of work, Rosch's colleagues at the University of California at Berkeley,
anthropologists Brent Berlin and Paul Kay (1969), showed that the color
terms used by diverse societies all follow a systematic pattern. That is, if
a society has only two color terms, those terms will divide the spectrum
between white and black; if a language contains three terms, then it con-
tains a term for red; if it contains four terms, then it contains a term for
green or yellow (but not both) and so on to the most complete languages
which contain eleven basic color terms (white, black, red, green, yellow,
blue, brown, pink, purple, orange, and gray). While these lines of work
have engendered controversy (as we shall see in chapter 12), they have
helped swing the pendulum of anthropological analysis back to the pole
of universalism: most researchers now believe that individuals the world
over perceive and classify in relatively similar ways, and that the ways in
which they classify reflect the operation of deep principles of mind which
cannot easily be dislocated.
Committed anthropologists thus face a dilemma. On the one hand, as
enemies of racism and cultural chauvinism, they are delighted by the
evidence that individuals the world over appear to think and process
information in similar ways. One of the fundamental puzzles of anthropol-
ogy appears to have been resolved. On the other hand, as scholars who
value the peculiar profiles of different cultures, they do not want these
reassuring signs of universalism to invalidate or render superfluous their
careful study of individual cultures. Thus, they take care to stress that the
identity of mental processes should not in any way lessen the importance
of chronicling and then explaining the vast differences around the globe
in behavior, in patterns of thought, and in the uses to which both are put.
255
One productive way in which the anthropological community can

have its cake and eat it too is to continue the careful case studies that have
been the lifeblood of the field, but to inform these studies with promising
concepts or methods from cognitive science. In this vein a number of
careful studies have been carried out by Michael Cole and his associates.
For example, Jean Lave (1977) has contrasted the mathematical skills ex-
emplified by Liberian tailors within their ordinary working context with
their performances on standard kinds of mathematical tasks. Sylvia
Scribner (1984) has documented that dairymen at work in the New York
area can carry out complex forms of numerical estimation while at work
in their customary environments. Edwin Hutchins (1980) has shown that
individuals engaged in debates about land rights in the Trobriand Islands
exhibit logical reasoning of considerable complexity. And Cole, working
with Sylvia Scribner (Scribner and Cole 1981), has studied the effects of
different varieties of literacy on the general reasoning and mnemonic ca-
pacities of the Vai group in Liberia. All these efforts cherish the individual
details of particular groups in their home context: they pointedly spurn
premature generalizations or excessive reliance on arbitrary sorting tasks.
Still, when proper caution is taken, forms of thinking in remote settings
do lend themselves to comparison with the kinds of thought process ex-
hibited and the kinds of measure used in traditional Western-schooled
settings. Anthropology can, indeed, be infiltrated by cognitive scientific
concepts, without any need for a wholesale abandonment of traditional
methods or classical problems.
One final point of contact between anthropology and the rest of
cognitive science should be noted. While most energy in recent years has
been devoted to the increasingly fine-grained analysis of particular do-
mains-ranging from kin terms to folk theories of emotions-some inves-
tigators remain interested in the general conundrums dating back to Tylor,
of how culture is possible, how it is constituted, and how it is acquired.
Part of the interest unfolds in an evolutionary framework. Investigators
working with other primates, with the skulls of earlier humans, or with
other relics of the paleontological or archaeological record seek clues about
how human beings became cultural animals-about the sources of lan-
guage, of social behavior, of rituals and art. Moving to a briefer time frame,
other investigators have raised the question of how children in a society
"learn" culture. In Roy D'Andrade's formulation (1981), culture consists
of a large pool of information passed from one generation to another,
including learned programs for action and understanding. Yet unlike the
classical computer programs, these programs tend to be unspecified and
inexplicit, absorbed through a slow process of guided discovery, and in-
volving the manipulation of content-centered rather than formal-based
256
symbolic systems. Every individual must acquire hundreds of thousands
of chunks of cultural information to be competent in the business of
society, and yet the mechanisms whereby these bits are acquired remain
obscure. D'Andrade does venture his opinion that culture is unlikely to be
picked up without considerable modeling and guidance, but that the way
this instruction takes place has not been illuminated thus far by standard
psychological accounts. Here, again, the model of how individuals learn
language presents itself as a tempting, if possibly perilous analogy.
Levy-Bruhl Revisited
In the century since the term culture became formally introduced, and the
discipline of anthropology was first formulated, massive amounts of em-
pirical information have been obtained about individuals in different cul-
tures. We know much more than earlier observers about kin terminology,
social organization, modes of behavior, ways of classification, use of lan-
guage and myth, powers of reasoning. As in many other areas of cognitive
science, the basic scientific dilemma-Do primitives think the way we do?
-has not lost its urgency. But the ways in which scholars think about this
question have become much more sophisticated. Nowadays most inves-
tigators take for granted that the basic modes of perception and classifica-
tion are the same everywhere, but that particular elements in the environ-
ment can affect how-and the extent to which-these processes develop.
Thanks to research in this tradition, many instructive differences among
individuals in diverse societies have been uncovered, even as the funda-
mental continuities in mental processes everywhere seem increasingly to
be confirmed. Here is where future work is likely to proceed.
As anthropologists have tackled the issues vividly raised in Levy-
Bruhl's own writings, they have found themselves engaged in an odd
assortment of activities. There have been flirtations with aggressively cog-
nitive measures-the formal musings of Levi-Strauss and the more pub-
licly verifiable componential analysis of the American school. While nei-
ther has proved a panacea for explicating the thought processes of any
group or individual, let alone an exotic one, clear insights have been gained
about at least some domains in some societies. Thus, Levi-Strauss has
tackled the broadest questions, but has failed to provide sufficiently rigor-
ous descriptions of his analytic methods; in contrast, the componential
analysts have developed relatively precise analytic tools but have not
successfully applied them beyond certain relatively restricted domains.
257
The success of the cognitive scientific approach to anthropology will

hinge on whether the rigor of componential analysis (or some other com-
putationally inspired approach) can be wedded to the broad issues that
have traditionally attracted scholars to the study of exotic cultures. We
have seen that the componential approach achieved reasonable (though
not unmixed) success in dealing with constrained categories like kinship
or color but has not so far been particularly revealing with broader catego-
ries such as emotions or diseases.
In a sense, it is useful to think of anthropology as representing a kind
of "upper bound" for cognitive science. Anthropology clearly deals with
issues representing very large bodies (such as entire cultures) and spanning
a quite wide scope (such as the relationship between a culture's linguistic
practices and its thought patterns). If cognitive-scientific methods can deal
successfully with such global issues, they will clearly have established
their utility in illuminating human thought.
It may turn out, however, that cognitive scientific methods are only
partially successful in dealing with such a broad assignment or can only
be usefully brought to bear upon the most constrained (and possibly the
least interesting) domains. While issues of context, culture, and history can
be bracketed in other cognitive sciences, they may be of the essence in
anthropology. If, in the last analysis, anthropology proves to lie largely
outside of the mainstream of cognitive science, this will be an important
(if somewhat disappointing) finding. And it may signal the even less happy
outcome that large areas of psychology, philosophy, and linguistics may
also fall outside of cognitive science, at least as currently practiced.
In the years since Levy-Bruhl debated with himself about the status
of the primitive mind, the pendulum has swung decisively to a belief in
the essential unity of the human mind everywhere. But the question of how
to study that mind remains hotly debated. For every believer in empirical
cognitive methodology, for every supporter of the structuralist or ethnose-
mantic approach, there is now a Geertzian, who believes that cognitive-
science approaches are essentially inadequate in the anthropological realm,
and who calls instead for a sympathetic reading of the cultural data, much
in the manner of a literary critic. It is worth noting, however, that in the
last several years a moderate middle ground seems to be emerging. Accord-
ing to this tack, anthropology remains the field where careful case studies
are indispensable and where keen attention to particularities remains of
enduring importance. At the same time, there is no reason why these
studies cannot be informed by the most salient and useful cognitive con-
cepts and analytic frameworks. And thus a new cadre of workers has
learned to speak the language and to carry out the approaches of cognitive
science, even though such researchers practice it in an area where the
258
specifics remain all important. Cognitive science can contribute to anthro-
pology, without enveloping it.
While the historical and particularistic aspects of each society con-
tinue to be valued, there is fresh recognition that, in the last analysis, all
that can be attained by any individual in any culture is restricted by the
particular species to which one belongs, and, more specifically, by the
nervous system that one possesses by virtue of one's humanity. For this
reason, anthropologists of every stripe-from Levi-Strauss to Geertz-
have savored discoveries about the human as an organism: the evolution
of the brain, the development of the skeletal musculature system, the
nature of sexual ties at different ages. Such insights from the areas of
biology and neuroscience will not in themselves answer questions about
culture-the levels of analysis are simply too disparate. (Indeed, neuro-
science serves as a kind of "lower bound" to cognitive science and thus is
maximally distant from anthropology.) But, in due course, findings from
the study of the human nervous system may well illuminate how an
individual becomes able in such short order to assimilate and to transmit
to others the practices of the culture in which he or she happens to live.
259
9
Neuroscience:
The Flirtation with
Reductionism
Karl Lashley Poses a Research Agenda
Karl Lashley, the pre-eminent American neuropsychologist of the first half

of the twentieth century, liked to recall his initiation into the intricacies
of the nervous system. As a seventeen-year-old laboratory aide in zoology
in 1907, "I found in a box of trash ... a series of Golgi stained sections
of a frog's brain. I proposed ... that I work out all the connections among
the cells, so that we might know how the frog works .... I have never
escaped from the problem" (quoted in Beach et al. 1960, p. xvii).
True to this testimony, Karl Lashley devoted his research career to an
investigation of the nervous system. Some of his early work was conducted
with J. B. Watson, who was shortly to found behaviorist psychology; but
the main influence on his scientific development was Shepherd Ivory
Franz, a neuroanatomist who was skeptical about the possibility of localiz-
ing behavior in specific regions of the nervous system. Among his principal
discoveries, Franz had found that a lesion in the frontal lobes of mammals
does not abolish learned behavior unless the destruction of the tissue is
massive; that long-established habits tend to persist in any case; and that
habits lost by extensive destruction can be relearned. Under Franz's tute-
lage, Lashley began a lifetime's effort to discover the neural substrate of
particular behaviors (Boring 1950).
260
Neuroscience: The Flirtation with Reductionism
The Lesion Technique
In carrying out his research, Lashley made extensive use of the tech-
nique of ablation, where specific areas of the nervous system (often regions
of the cerebral cortex) are destroyed by means of a surgical lesion. The
basic goal of the ablation technique is to determine which behavior is
impaired or destroyed following a punctate lesion, and thereby to infer
which functions are typically served by that region of the brain. Lashley
was inclined to be skeptical about the possibility of attributing specific
behavior to specific regions of the brain; but in no way did the apprentice-
ship with Franz foreshadow the extent to which Lashley would go to
pursue this problem or the bold conclusions he was to reach some decades
later. The experiences of the student cannot foretell the achievements of
the mature scientist.
In the decade following his initial training by Franz, Lashley con-
ducted dozens of experiments on the nervous system of the rat. His typical
study involved ablation of an area of the visual cortex of the rat brain and
a determination of the effects wrought on the rat's perceptual powers. By
1929 Lashley was ready to summarize his findings in a major work entitled
Brain Mechanisms and Intelligence. He strongly questioned the significance of
specific neural zones and connections: "It is very doubtful that the same
neurons or synapses are involved even in two similar reactions for the same
stimulus" (1929, p. 3). Describing experiments on maze running after
cortical ablation was examined, Lashley concluded:
The capacity to learn the maze is dependent upon the amount of functional
cortical tissue and not upon its anatomical specialization. . . . The results are
incompatible with theories of learning by changes in synaptic structure, or with
any theories which assume that particular neural integrations are dependent upon
definite anatomical paths specialized for them .... The mechanisms of integration
are to be sought in the dynamic relations among the parts of the nervous system
rather than in details of structural differentiation. (P. 3)
With these words, Lashley hurled a set of sharp challenges to the

neuroscientific community. After all, a principal lure of neuroscience is the
hope that the specific neural bases of particular behavior can be found-
and what better prize than discovery of the neural focus of a particular act,
thought, or sequence of behavior? Yet here was a major neuroscientific
researcher declaring, on the basis of decades of research, that such a search
was forlorn. Lashley was calling into question localization-the belief that
specific behavior resides in specific neural locations. At the same time, if
less explicitly, he was also posing difficulties for reductionism, the scientific
program that seeks to explain behavior entirely in terms of neural (or other
lower-order) principles.
261
II I THE CoGNITIVE SciENcEs : HisTORICAL PERSPECTIVE
Equipotentiality and Engrams

Having established to his satisfaction that particular behavioral pat-
terns cannot be consigned to specific cortical regions, Lashley spent the rest
of his scholarly life searching for a viable alternative to the classic localiza-
tionist position (Beach et al. 1960). He was attracted to the principal ideas
of Gestalt psychology, according to which the organism perceives overall
patterns initially and only subsequently becomes sensitive to their compo-
nent parts. From Lashley's perspective, this approach to psychology made
neurological sense as well: perhaps, rather than responding to specific
forms of information, the brain works as an integrated unit, responding as
an organized totality to complex patterns of stimulation.
Lashley developed several related concepts. He spoke of equipoten-
h'ality-the capacity of any part of a functional area to carry out a particular
behavior. Equipotentiality is mediated by the law of mass action, according
to which the efficiency of performance of a function may be reduced in
proportion to the extent of the brain injury within an area. In other words,
impairment in performance is due not to the site of the injury, but rather
to the amount of tissue destroyed. As Lashley put it:
The alternative to the theory of the preservation of memories by some local

synaptic change is the postulate that the neurons are somehow sensitized to react
to patterns or combinations of excitation. It is only by such permutations that the
limited number of neurons can produce the variety of functions that they carry out.
. . . All of the cells of the brain are constantly active and are participating, by a sort
of algebraic summation, in every activity. There are no special cells reserved for
special memories. {1950, p . .ri)
Finally, he pondered the property of plasticity-the potential for remaining

areas of the nervous system to take over when a specific region has been
damaged.
Given this belief in generalized patterns of representation and behav-
ior, what was one to make of the search for the engram, the discrete repre-
sentation in the nervous system of specific ideas, concepts, or behaviors?
Lashley came to the conclusion that one would never find such an engram.
As he put it in a provocatively negative formulation, "This series of experi-
ments has yielded a good bit of information about what and where the
memory trace is not. It has discovered nothing directly of the real nature
of the engram. I sometimes feel, in reviewing the evidence on the localiza-
tion of the memory trace, that the necessary conclusion is that learning just
is not possible" (1950, p. 501). In Lashley's view, during learning, informa-
tion comes to be represented widely within large regions of the brain, if
not throughout the brain as a whole. Within these areas, all the cells
262
Neuroscience: 7he Flirtation with Reductionism
acquire the capacity to react in certain definite patterns. Whether the cells
can be mobilized to carry out an impaired function depends upon the
percentage of them still remaining after brain injury, the degree to which
the pattern of behavior has been mastered beforehand, and the strength
of motivation of the animal. Particularly in the case of brain-injured ani-
mals, motivation might have to be quite potent if the organism were to
display its still-preserved abilities.
Lashley's experiments strike a severe blow against any unmodulated
account of the nervous system as a set of discrete centers, each having its
own unique functions, so that destruction of a set of cells would result in
complete loss of a particular function and in total sparing of all remaining
ones. At the same time, he helped to cast doubt on the reflex arc-the bond
whereby each response is triggered by a specific stimulus-which had been
the principal neural model of behavior in higher (as well as lower) organ-
isms. Things were simply not that simple. In study after study, Lashley
deduced the implications of the localizationist position, designed a sup-
posedly critical experiment, and showed that abilities persist despite the
predictions of the localizers, thus giving the lie to those scientists who
claimed that impulses have to be transmitted over certain established paths
in order for specific behavior to be performed.
s
Lashley Iconoclasm
In addition to providing a devastating critique of a simple-minded
localizing position, Lashley also impressed on the next generation of work-
ers the difficulty of coming up with a viable model of the nervous system.
In possibly his best-known paper, the Hixon contribution "The Problem
of Serial Order in Behavior" (Jeffress 1951), which I discussed in chapter
2, he evocatively laid out a whole set of problems that neurobiology had
ignored. Drawing on examples from language, walkin& playing the piano,
and the like, Lashley demonstrated that many sequences of behavior ex-
hibited long planning units which unfold too quickly for them to be
altered or corrected "live." In his view, it was necessary to reconceptualize
current associationist models of the nervous system to allow for effects
that can be manifest for a significant period after initial stimulation. And
so, to choose a single example, in order to understand a double entendre, one
would have to retain in mind (and in the brain) a meaning latent in the
body of the joke, which only becomes "activated" by the punch line. No
simple stimulus-response bonds can explain this behavior: one needs a
model of the nervous system which is hierarchically arranged and features
feedback and feed-forward mechanisms.
Lashley's unorthodoxy manifested itself in the new talk about com-
263
puters. At the time of the Hixon Symposium, many noted scholars were
eager to analogize the brain to the digital computer. Lashley voiced a note
of caution:
The brain has been compared to a digital computer because the neuron, like
a switch or valve, either does or does not complete a circuit. But at that point the
similarity ends. The switch in the digital computer is constant in its effect, and its
effect is large in proportion to the total output of the machine. The effect produced
by the neuron varies with its recovery from [the] refractory phase and with its
metabolic state. The number of neurons involved in any action runs into millions
so that the influence of any one is negligible .... Any cell in the system can be
dispensed with .... The brain is an analogical machine, not digital. Analysis of its
integrative activities will probably have to be in statistical terms. (Quoted in Beach
et al. 1960, p. 539)
By his careful experimental example and his polemical style of expression,

Lashley called into question central dogmas of the time. He was saying to
his colleagues that their vaunted stimulus-response model, their reflex arc,
cannot explain behavior. As his student Karl Pribram recalls, "Lashley had
a genius for skepticism, for poking holes in all the myths that had grown
up about how the brain works" (Hooper 1982, p. 170).
Less directly, but with equal decisiveness, Lashley's descriptions un-
dermined unreflective assumptions about reductionism. So long as ac-
counts were based upon specific cells or particular reflex arcs, neuroscien-
tists could rest easy in their belief that behavior could be explained at the
neurological level. A particular lobe controls vision or subserved volition,
and that was that. On this assumption, the need for explanation on the
psychological or mentalistic level might fade away. But ever the scholarly
troublemaker, Lashley spent the last years of his professional life describ-
ing behavior-for example, long sequences carried out without feedback
-that eluded current mechanistic models and strongly implied more ab-
stract and hierarchically organized forms of representation. While Lashley
himself did not call for explanations on the mentalistic level-he was still
too much of a Watsonian-his work (and his talk of "plans" and "struc-
tures") cleared the way for Simon's belief in a symbol system, Piaget's call
for mental operations, Miller's TOTE system, and Chomsky's resorting to
rules and representations. In our terms, Lashley helped to set the stage for
a cognitive-scientific approach to behavior and thought.
While Lashley's substantive contributions to cognitive science should
not be minimized, his actual claims within neuroscience met a less positive
fate. As has happened with other workers who put forth strong and falsifi-
able programs, succeeding researchers have put many of his claims to the
test. Today few of his specific propositions would still command wide
264
assent. For instance, the nervous system turns out to be far more specific,
far less equipotential than Lashley had contended. His belief that the brain
works in a Gestalt-like fashion would find few adherents today. But in
framing the questions sharply, and in introducing the key terms of scien-
tific debate, Lashley for a long time dominated, and still exerts appreciable
influence on, that work in neuroscience that impinges upon cognitive and
behavioral concerns.
The work of Lashley, and of other experimental researchers working
at the intersection of the brain and behavior, raises another issue-and one
crucial for my inquiry here. It has been held by many scientists, especially
neuroscientists, that the optimal way to account for human behavior and
thoughts is in terms of the structure and functioning of the human nervous
system. To some investigators, this neuroanatomical account can comple-
ment accounts proffered in psychological or behavioral language; but for
others, neuroanatomical accounts may eventually render unnecessary ac-
counts in terms of representations, or symbols, or other psychological
argot. In the view of this latter reductionist group, cognitive science
emerges as, at best, a holding operation: a temporary account of mental
activity destined to vanish once an account in terms of synapses can be
attained. And such reductionists find themselves in sharpest opposition to
functionalists-true-blue cognitivists who believe that behavior and
thought must be accounted for completely on the level of representations,
without any regard whatever to the "hardware" in which it happens to be
embodied. The debate about the possibility-and the desirability-of re-
ductionism lurks in the background in any account of neuroscientific work
(see Mehler, Morton, and Jusczyk 1984).
How Specific Is Neural Functioning?
Debates about the specificity of function go back a long time in scientific

history. Descartes, among the first to concern himself with the relation-
ship between the body and mind, put forth his own notions of localiza-
tion. He recognized that different parts of the brain control different
bodily functions, and located the interaction between the soul and the
body at the pineal gland at the base of the brain. Juan Huarte, a contem-
porary of Descartes, rejected the doctrine of the localization of faculties
in separate ventricles, and suggested instead that the brain works as a
unit (Diamond 1974). Even at the dawn of the scientific era, it seems,
dispute was already raging about the plausibility of the reductionist po-
265
sition and the viability of a holistic (as against a localizationist) account

of neural representation.
It is not surprising that scientists have long been interested in the issue
of how processes and information are represented in the brain. Unless one
subscribes to a wholly mystical account of human behavior-in which case
one is not a scientist-it becomes important to try to understand the bases
of human behavior and thought. Even the Greeks put forth theories about
this issue, though the particular accounts they wove strike us today as
more poetic than plausible. Once the brain had been selected as the region
of the human body most likely to be involved in matters of mind, discus-
sion about the neural basis of thoughts and actions necessarily came to
center on this circumscribed region.
Understanding of the nervous system developed rapidly in the eigh-
teenth century (see Gardner 1975; Herrnstein and Boring 1965). Luigi
Galvani showed that electrical charges can cause muscles to contract.
Charles Bell and Franc;ois Magendie independently discovered the
anatomical separation of sensory and motor functions in the spinal cord.
Animals could continue to move even after being anesthetized, just as
animals could still undergo sensation after being paralyzed. Johannes
Muller put forth the law of specific energies, which held that the quality of
experience is determined not by the features of the objective stimulus but
rather by the particular neurons that respond to it.
Building on such discoveries, Francis Joseph Gall was able to pro-
pound the most far-reaching (and most notorious) theory of localization.
The title of one of his works summarizes his major claim: "On the func-
tions of the brain and each of its parts: with observations on the possibility
of determining the instincts, propensities, and talents, or the moral and
intellectual dispositions of men and animals, by the configurations of the
brain and the head." In his writings, Gall defended the proposition that the
brain is so divided as to be able to carry out many discrete functions. These
dispositions are innate, and "the brain is composed of as many particular
and independent organs as there are fundamental powers of the mind"
(quoted in Robinson 1976, p. 339). In the hands of some of his more
extreme followers, Gall's doctrine of phrenology led to the claim that one
could even discern an individual's unique intellectual profile by examining
the specific configurations of his skull.
While appealing widely to the public of the early nineteenth century,
Gall's claims did not go over well in the scientific community. His chief
critic, Pierre Jean Marie Flourens, carried out experiments that demon-
strated common actions of different parts of the brain. Flourens insisted
that the cerebral regions function as a whole and that the magnitude of a
deficit cannot simply be reduced to the area or even the amount of brain
266
Neuroscience: The Rirtation with Reductionism
involved. While conceding some specificity in the nervous system, Flour-
ens added that "there should, moreover, properly be brought to light
another order of phenomena that includes both this efficacious Unity of the
nervous system that joins all the parts of the system together, in spite of
their diversity of action, and also the degree of influence that each of
these parts contributes to the common activity" (quoted in Hermstein and
Boring 1965, p. 222).
Evidence for Localization

Toward the end of the nineteenth century, the pendulum began to
swing from Flourens toward Gall. Gustav Theodor Fritsch and Eduard
Hitzig (1870), two German neurophysiologists, were able to demonstrate
unequivocally that stimulation of different areas on the cortex of a dog was
followed by the contraction of specific muscles. David Ferrier, working in
England a few years later, directed his attention to the prefrontal areas of
the cortex. Though neither motor nor sensory functions were demonstra-
ble in this region, Ferrier maintained that
[it) is important for intellectual work. ... [Following ablation] I could perceive
a very decided alteration in the animal's character and behavior .... Instead of as
before being actively interested in their surroundings and curiously prying into all
that came within the field of their observation, they remained apathetic .... They
had lost, to all appearance, the faculty of attentive and intelligent observation.
(Quoted in Diamond 1974, p. 244)
Soon such localizations of higher cortical functions were also being

discovered in human beings. In the 1860s, Paul Broca (1861), a French
surgeon, reported the cases of two aphasic patients, individuals who had
lost language as a result of injury to the brain. In a finding whose impor-
tance for subsequent scientific inquiry is difficult to overestimate, Broca
claimed that this damage to language had not simply resulted from random
injury to the nervous system but was restricted to insults to the left
cerebral hemisphere. Narrowing his focus even further, he claimed a spe-
cial priority for expressive language, in the lower portion of the third
convolution of the frontal lobe. Subsequent case reports confirmed the gist
of Broca's claim and also suggested an association of other kinds of linguis-
tic breakdown with injuries to related sites in the left hemisphere. Just a
few years later, Carl Wernicke (1874) traced difficulties in understanding
language to an injury in the left temporal lobe; while Jules Dejerine (1892)
attributed difficulties in reading and writing to lesions in the left parietal
lobe and the parietal-occipital cortex (see Gardner 1975).
It was a heyday for scholars of a localizationist persuasion. With
267
increasingly sophisticated methods for testing animals, claims were made

for specificity in each region of the cortex. In a parallel fashion, as addi-
tional case studies of brain-injured patients accumulated, claims about
the astonishing specificity of certain cognitive deficits were forth-
coming. Perhaps-the dream went-if studies of sufficient particularity
were made, the functional significance of all areas of the brain might be
laid bare. Localizationism would triumph; reductionism would gain in
plausibility.
The followers of Flourens, skeptical about localization claims, were
not about to give up the game. A few decades after Broca's epoch-making
publication, Pierre Marie, a French neurologist, declared to the Paris
Neurological Society that "the third frontal convolution plays no special
role in the function of language" (1906, p. 241). Marie had re-examined
the brains originally studied by Broca and come to the conclusion that the
master's original claims were simply not justified. Each of Broca's patients
had far more extensive lesions than Broca had reported, and the range of
accompanying deficits had not been documented with sufficient precision.
Marie countered that there was but one form of aphasia, which could arise
from lesions in various cortical areas; reports of localization had been
greatly exaggerated.
The Resurgence of Holism

Marie's protest might have been enunciated in a void; but, in fact, it
generated many echoes. Within a few years, a variety of neurologists had
endorsed his claims that cognitive functions are not highly localized in
the nervous system. They adduced evidence that the same kinds of defi-
cits could be obtained from individuals with lesions in a wide variety of
areas; and conversely, patients with similar anatomical lesions often ex-
hibited contrasting sets of deficits or even at times no deficits at all.
These neurologists, who came to be known as the "holists" (in contrast
to the "localizers" or "diagram makers") regarded the brain as a single
highly integrated organ, one involved as a whole in all intellectual activi-
ties and not susceptible to specific impairment from discrete lesions.
They spoke of the plasticity of the nervous system, the capacity of unin-
jured areas to take over from injured areas, and the loss of abstract
thinking and other functions as a consequence of the size, rather than the
site, of lesion (Gardner 1975).
The holist position assumed by neurologists like Pierre Marie, Kurt
Goldstein, and Henry Head was reinforced by certain events within the
field of psychology. Like Karl Lashley, their counterpart in the world of
animal neuropsychology, these investigators were much impressed by the
268
Neurosdence: The RirfaHon with ReducHonism
findings from Gestalt psychology. It was clear that the organism reacted
not just to single stimuli but rather to the relationship among stimuli, to
overall patterns, and to stimuli perceived as part of a given context. More-
over, the Gestalt assumption that the nervous system is organized in terms
of neural fields, operating across wide regions of the cortex, struck a
responsive chord with these neurologists. The opponents of localization in
neurology, like the opponents of atomism in psychology, served as a
bulwark against the elementarist bias of the behaviorists during the early
part of the century. At the same time, holists were far more sympathetic
to the notion that behavior could not be explained satisfactorily in terms
of neural circuitry. As they saw it, there was a continued need for explana-
tion on the psychological level-particularly the global or holistic kinds of
explanation offered by the Gestalt psychologists.
Once again, then, there was a correlation between skepticism about
localization and skepticism about reductionism-not a logically necessary
association, to be sure, but rather a meeting of two ideas in the minds of
many scientists. In the view of these investigators, even if it could be
shown that certain functions regularly break down as a result of damage
to specific areas of the brain, the significance of such a finding was not
clear. Hughlings Jackson, a forerunner of the holist school during the
nineteenth century, had declared that localization of symptoms did not
signify localization of function (Jackson 1932). To give a specific exam-
ple, just because naming breaks down following lesion in the angular
gyrus, it is an unwarranted assumption that naming actually takes place
in this specific site of the brain. Richard L. Gregory has stated this point
crisply:
The removal of any of several resistances in a radio set may cause the emission
of strange sounds, but it cannot therefore be concluded that the function of the
resistances is to inhibit howling .... The southern region of British railways is a
complex system of railway lines, signal boxes, stations and control systems. A
breakdown of a section of the line, a power failure or a slip in the central control
room at Waterloo, may disrupt traffic over a wide area. But we cannot therefore
say that the function of the system is localised in the ... power station, the central
control room ... all are essential. (Quoted in Rose 1973, p. 94)
We have here what looks like a scientifically untenable situation. One

school of investigators, represented by the localizers, was claiming increas-
ingly fine kinds of informational specificity in the nervous system. From
their perspective, it was only a matter of time before every behavioral
function could be adequately mapped in the brain of the organism. The
rival school of investigators, equally respectable and equally·vocal, consid-
ered the localizationist approach to be bankrupt: not only was evidence
269
mounting in favor of the trio of mass action, equipotentiality, and plas-

ticity; but even to the extent that the localizers could demonstrate links
between lesions and deficits, the significance of this association remained
unclear.
Evaluating the Evidence

When respectable groups of scientists have such opposite points of
view, it is necessary to step back and attempt to gain some perspective. It
is possible, of course, that one school is entirely correct and the other
entirely wrong; but neither school dismissed the experimental claims of the
other, so this denouement was not likely. It is also possible that each school
was talking about different phenomena and was therefore correct in its
own bailiwick. Thus, perhaps the localizers were correct for certain organ-
isms, for certain behavior, for certain portions of the life cycle; while the
holists were correct for other organisms, other behavior, or other periods
of life. And, indeed, each school of investigators did look to somewhat
different clusters of evidence in order to bolster its conclusions. A holist
typically turned to maze-running rats, while a localizer was likely to gravi-
tate toward the varieties of aphasia in adults.
At any rate, by the end of the 1940s, many investigators were seeking
some rapprochement between the rabid holists and the extreme localizers.
It was becoming clear that claims about highly specific syndromes follow-
ing highly specific lesions could not be maintained; the variation across
patterns, and across clinics, was simply too great. On the other hand, any
number of lines of investigation undercut the extreme holist position. Even
Lashley himself conceded that there was a fair degree of localization within
the visual system-the system he happened to know the best. Studies by
neuroembryologists like Paul Weiss (1952) and Roger Sperry (1951) re-
vealed that the developing neurons contain highly specific information:
when budding limbs were surgically transplanted to a new position, the
nerves would still reconnect with the place of origin, even if it meant that
the organism was unable to move toward its usual targets. Such findings
challenged extreme versions of the plasticity hypothesis and demonstrated
considerable specificity and committedness even in the immature nervous
system. Most dramatically, the Gestaltist claims about "fields of brain
representation" were undercut by experiments in which various encum-
brances were placed in the brain itself. Lashley placed gold foil to disrupt
neuroelectric fields (Lashley, Chow, and Semmes 1951); Sperry inserted
insulated mica strips into the brain (Sperry and Miner 1955); and Karl
Pribram (1971) put aluminum hydroxide cream in minute amounts over
the cortical surface. Pattern discrimination remained intact despite marked
270
Neuroscience: The Rirfation with Reductionism
disruption of direct currents and electroencephalographic activities. These
results were difficult to explain on a localizing account; yet, at the same
time, they discredited the major theoretical account put forth by the Ge-
stalt holists. Faced with the results of these experiments, Gestaltist Wolf-
gang Kohler, the chief proponent of field theory, is said to have declared
in desperation "that ruins not only my Direct Current field (theory] but
every other current neurological theory of perception" (quoted in Pribram
1971, pp. 110-11).
Donald Hebb's Bold Synthesis
A theoretical synthesis is at a premium when the competition between

theories becomes too strident. Probably the person who most successfully
mediated between the Gestalt and the atomist-localizing points of view
was the Canadian neuropsychologist Donald 0. Hebb. In his now classic
monograph The Organization of Behavior (1949), Hebb argued that behavioral
patterns, such as visual perceptions, are built up gradually over long peri-
ods of time through the connection of particular sets of cells, which he
called cell assemblies. To this extent, behaviors or percepts can indeed be
localized in specific regions, perhaps even in specific cells of the brain.
However, with time, more complex behaviors come to be formed out of
sets of cell assemblies, which he called phase sequences. These phase se-
quences are less localized, and involve much larger sets of cells drawn from
disparate sections of the nervous system. A phase sequence inevitably
involves some equipotentiality: it includes alternative pathways so that if
some are destroyed, a behavioral function can still be carried out with
greater or lesser effectiveness by those that are spared. Finally, by the time
the organism has reached maturity and is capable of performing the most
complex forms of behavior, it is difficult to attribute any behavior to a
discrete set of neurons in a delineated region of the brain.
One can see how Hebb's position could comfortably accommodate
portions of both the holist and the localization points of view. On the one
hand, early in life, the localizer can be seen to have the edge, since simple
perceptions depend on specific sets of cells: no holism here. Yet, with
development, more complex cell assemblies and phase sequences are
formed, and these are able to participate in numerous kinds of behavior:
plenty of holism here. Still, it would be an oversimplification to see the
developmental course as proceeding from localizing to holism; for, in other
respects, the sequence is exactly the reverse. As Hebb pointed out, learning
271
early in life is flexible, so that it can take place despite the destruction of
large parts of the nervous system; in contrast, later learning depends spe-
cifically on certain developed structures, and there is relatively little plas-
ticity in the system. In this respect, then, it is more plausible to view the
developmental sequence as running from holism to localization. A benefi-
cial effect of Hebb's work was to point up these various complexities and
competing tendencies, making it less plausible for anyone to adopt a rigid
localizing or an inflexible holist position. This intermediary position came
to carry considerable weight and has recently been embraced by computer
scientists who are attempting to simulate vision (see chapter 10).
The Hixon Symposium Revisited
The need to reconcile these extreme positions was also manifest at the
various pivotal conferences during the late 1940s and early 1950s. During
the Hixon Symposium (Jeffress 1951), the major neurologists, neurophysi-
ologist&, and neuropsychologist& debated with one another the tenability
of the localization position. Psychologist Heinrich Kluver reported aston-
ishingly specific behavior consequent upon lesions in the temporal lobes
of monkeys: these organisms exhibited psychic blindness, strong oral fixa-
tions, an excessive tendency to react to each and every visual stimulus,
profound changes in emotional behavior, a remarkable alteration in dietary
habits, and an increase in the amount and diversity of sexual behavior.
Neurologist J. M. Nielsen described specific forms of agnosia (failure to
recognize objects) which accompany lesions in the occipital lobes. Neuro-
psychologist Ward Halstead discerned predictable mental difficulties in
planning and in abstraction following damage to the frontal lobes.
In response to these strands of evidence in favor of localization, Karl
Lashley reviewed his own long series of experiments on pattern recogni-
tion and maze running in rats. Pointing out that he had found a remarkable
lack of evidence for localization, he taunted Halstead:
In fantasy, I have thought perhaps that my most important contribution when

I reach retirement age would be to have my frontal lobes removed and see what
I could do without them. I have less confidence than Dr. Halstead that it would
preclude the production of something of interest. We have little experimental
evidence of intellectual defect from uncomplicated removal of the prefrontal lobes.
(Quoted in Jeffress 1951, p. 145)
Wolfgang Kohler voiced his own skepticism about localizationalist

accounts: "The atomistic character of Dr. McCulloch's neurophysiology
272
Neurosdence: The Flirtation with Reductionism
prevents any direct approach to relationally determined facts such as visual
shapes. . . . If we think of cortical function in terms of continuous field
physics rather than of impulses in neurons, the difficulty never arises"
(quoted in Jeffress 1951, p. 65). Such, then, was the situation in the early
1950s: Kohler and Lashley defending a top-down perspective, against
proponents of a view based on specific neural structures. Lines were still
drawn between these two factions, along with a growing feeling that each
of them must hold part of the truth, and a widely discerned need for a more
integrative point of view (like Hebb's) which could give each position its
due.
Hubel and Wiesel's Decisive Demonstrations
In the late 1950s, David Hubel and Torsten Wiesel, two young neuro-
physiologists, began to record with microelectrodes from single cells in the
cortex of the cat. Over the next two decades, they were to record impulses
from the nervous systems of numerous cats and other animals as well, and
to do so throughout the visual system at various depths and in other
regions of the brain. For this work, widely recognized as pathbreaking,
Hubel and Wiesel were awarded the Nobel Prize for Medicine or Physiol-
ogy in 1981. Much of our current knowledge of localizing has been ob-
tained from this line of study (Hubel and Wiesel1959, 1962, 1979; Hubel
1979; see also Lettvin et al. 1959).
Hubel, Wiesel, and their associates documented two important
phenomena. First of all, they demonstrated beyond any doubt that specific
cells in the visual cortex respond to specific forms of information in the
environment. The exquisitely organized cerebral cortex contains cortical
columns, and simple cells within the columns respond to such punctate
properties of stimuli as orientation, or presence or absence of light. So-
called complex cells react to lines kept in optimal orientation as they sweep
across a receptive field; perhaps this mode of reaction corresponds to some
early stage in the brain's analysis of visual forms. Some cells respond to
input only from one or the other eye, while others can be influenced
independently by both eyes. For some hypercomplex cells, the most effective
boundary shape is a corner; for others, a tongue shape. Following such
demonstrations, no one could any longer doubt that the nervous system
was highly specific in its mode of functioning.
The second and equally important line of work emanating from the
Hubel-Wiesel laboratory concerns the critical role played by certain early
experiences in the development of the nervous system. While some per-
273
ceptual and motor abilities are clearly "wired in" at birth (or else how could
the organism function well enough to survive?), this is by no m~ans true
for all functions (or else, how could the organism change and learn?). The
visual system of the cat will not develop-in fact, parts of it will atrophy
-if the animal is not exposed to patterned light after birth. Moreover, the
cat must be exposed to a visually varied environment, permitted to use
both eyes, and allowed to move about its environment. If exposed to
horizontal patterns only, the cells normally destined to carry out vertical
processing will either atrophy or will be "taken over" to execute other
functions. It should be noted, too, that the timing of these early experi-
ences can be specific. For example, if, between the third and fifth postnatal
weeks, one of the eyes is prevented from seeing forms, it will become
functionally blind thereafter, even if normal registration and imaging are
restored.
While Hubel and Wiesel's work stands at the center of this line of
research, their efforts have been complemented and strengthened by those
of many other workers. Jerome Lettvin and his colleagues at Massachusetts
Institute of Technology showed early on that receptors in the eye of the
frog are also extremely specific and respond to small round bloblike shapes
which are rather reminiscent of bugs-hence, the term bug detector (Lettvin
et al. 1959). Vernon Mountcastle (1978) documented columns in the
somesthetic sensory cortex that have specific response patterns analogous
to those of the visual cortex. Recording from the inferior temporal lobes
of monkeys, Mortimer Mishkin (1967) finds cells that participate in coding
the physical attributes of visual objects; these cells appear to be involved
in activities much closer to the recognition of objects (and some distance
away from that sensitivity to punctate forms of stimulation which excite
the neurons in the primary visual cortex). There are even cells in the
monkey's cortex that respond maximally to the shape of a monkey's hand
(Gross, Rocha-Miranda, and Bender 1972). For those in sympathy with
specificity and localization of function, the last few decades have yielded
much confirming evidence.
The Molar Perspective
Though sets of cells responding to objects, or major portions of objects,

represent a more molar form of pattern recognition than do cells respond-
ing to spots of light or individual edges, they are still some distance
removed from the "higher" forms of thought which occupy most cognitive
274
Neuroscience: The Rirtation with Reductionism
scientists. Another set of studies carried out at a molar level also bears
upon the claims of the localizers and the holists.
Sperry on Split Brains

Because they were suffering from intractable epilepsy, a small group
of patients were subjected, in the late 1950s and early 1960s, to a radical
operative procedure: a surgical intervention where the two halves of the
brain were disconnected from one another. Roger Sperry grasped the pivo-
tal scientific implications of this unusual form of surgery and, with his
colleagues at the California Institute of Technology, devised methods for
testing separately the two halves of the brain. These studies were also of
epoch-making importance and resulted in Sperry's sharing the Nobel Prize
in 1981 with Hubel and Wiesel (Gazzaniga 1981; Sperry 1974; see also
Geschwind 1965).
The major thrust of Sperry's work was to document important differ-
ences in the functioning of two hemispheres. From a plethora of tests
carried out on the dozen patients available for scrutiny, Sperry reinforced
the impression of clinicians (Geschwind 1965) and experimentalists
(Kimura 1973; Milner 1967) that the left hemisphere is dominant for lan-
guage and other conceptual and classificatory functions, while the right
hemisphere assumes a dominant role for spatial functions and for other
fine-grained forms of discrimination. So far, then, support for the localiz-
ing point of view-and remarkable support, too: for each hemisphere of
the same person can be tested separately, in such a way that the other
hemisphere is not even "aware" of the testing going on with its neural
neighbor.
Yet, once again, a line of study also offered some support for the
holistic point of view. Through the use of ingenious techniques, the Sperry
team was able to show that the right hemisphere of right-handed persons
was capable of far more linguistic functioning than had hitherto been
thought: for example, there was considerable ability to understand written
language (Zaidel1977, 1978). Perhaps in the normal individual, the very
dominance of the left hemisphere may prevent documentation of the
latent linguistic capacities of the right (or nondominant) hemisphere. Sec-
ond and of equal importance, there were wide individual differences
among the commisurotomized patients (Gazzaniga 1983). In particular, the
younger they had been at the time of their operation, the more likely it
was that patients would reveal well-developed capacities in both hemi-
spheres. By contrast, those patients who sustained epilepsy and underwent
surgery relatively late in life exhibited the usual pattern of considerable
lateralization of function found in most brain-damaged patients. Finally,
275
the two hemispheres themselves differed, with the right hemisphere rela-
tively dominant for more holistic or Gestalt-forms of perception.
Gradients of Plastidty and Hierarchy of Functions

Sperry's results suggest that there is indeed considerable plasticity in
the immature human nervous system: the human being has the opportu-
nity-if not the necessity-of exploiting those portions of the nervous
system that under ordinary circumstances would not be mobilized for
these specific purposes. The earlier the time of trauma, the more likely that
one will prove able to carry out the desired function, irrespective of the
site of injury. Indeed, even if one loses an entire left hemisphere during
the first year or two of life, one will nonetheless be able to learn language;
at this time, the relevant zones of the right hemisphere are sufficiently
plastic or equipotential to assume this function. If the same operation were
to take place later in life, however-say, after adolescence-there would
be far less language recovery; in all likelihood, one would remain discerni-
bly aphasic or even mute thereafter (Dennis 1980).
Other factors also influence the degree of plasticity. In general,
younger individuals, those who are left-handed or who have sustained
some brain injury early in life also exhibit more plasticity than those who
exhibit contrasting traits. Thus, support for the localizing position is pro-
vided by data obtained with older right-handed persons; while that for the
holistic position is more likely to arise from studies of younger, left-
handed persons.
Studies conducted with brain-damaged patients in other laboratories
have also increased our understanding of the neural representation of
cognitive capacities (Geschwind 1974). From the work of Alexander Luria
(1966) and his colleagues in the Soviet Union, it is now clear that, with
development, different nervous centers gain dominance, and the hierarchy
among behavioral functions alters. For instance, in young children, sensory
regions are dominant; but in older individuals, the association cortexes and
the "planning regions" of the frontal lobes become ascendant. Thus, le-
sions of the sensory regions are more lethal in younger subjects, while
lesions of the frontal lobe are more pernicious in mature subjects.
According to Luria's studies, no function is carried out fully by a
specific region, but nor is it the case, as Lashley implied, that all regions
figure equally in a specific function. Rather, several anatomical regions
may figure in the performance of a particular behavior, but each of them makes
a characteristic and irreplaceable contribution. Thus, in the case of drawing, the left
hemisphere is responsible for the mastery of details, while the right hemi-
sphere contributes the overall sense of form. Hence, the kind of disruption
276
seen in an individual's drawings generally reveals the site of that person's
brain damage. We thus move from a simple assertion that one hemisphere
carries out one function and the other hemisphere a rival function, to the
more sophisticated claim that each hemisphere, or each region within a
given hemisphere, contributes to a given activity in characteristic ways.
Many sentences have been written about the optimal way in which
to describe the dominance patterns of the two hemispheres. Characteriza-
tions range from the mundane (language in the left, spatial function in the
right) to the grandiose (science or rationality in the left, art or intuition in
the right) to a distinction that captures current debates in neuroscience
(localized function in the left hemisphere, synthetic or holistic functioning
in the right). The precise characterization of the functions of the two
hemispheres-or, more probably, determination that such a neat dicho-
tomization is simply not possible--awaits the results of further studies
(Beaumont, Young, and McManus 1984). Yet whatever the mission of each
isolated hemisphere, there is a clearly dynamic interaction between the
two hemispheres. When the left hemisphere is aroused (for instance, by
the sounds of language), it promotes certain kinds of analytic and linguistic
functions. In contrast, arousal or stimulation of the right hemisphere
brings spatial and holistic functions to the fore (see Kinsbourne 1978).
While, in general, it is preferable to sustain an injury to the brain early
rather than late in life, and to exploit the plasticity of that developmental
stage, early is not always better. There are at least three caveats. First of
all, sometimes an early injury manifests no apparent deficits at the time but
produces severe deficits later in life. Thus, injuries to the frontal lobe in
a young monkey may produce no immediate impairment of functioning,
but long-term sequalae become evident at the time when the planning or
mnemonic functions of the frontal lobes would normally mature (Gold-
man-Rakic et al. 1982; Goldman and Galkin 1978). Second of all, when
another area of the brain takes over; the "rescuer" may well sacrifice the
potential for carrying out its own preordained functions. Thus, the right
hemisphere may assume language functions in the left-hemisphere-
injured child, but the child will eventually display spatial deficits, because
of the resulting unavailability of the relevant right-hemisphere zones (B.
T. Woods 1980). Finally, even when another area of the brain assumes a
function, it may not do so in an optimal way. Individuals with only a right
hemisphere do learn to speak, but they end up using different linguistic
strategies, which render them relatively insensitive to crucial syntactic
features of language (Dennis and Whitaker 1976). At least for right-hand-
ers, there may be only one optimal way to learn language-the way that
draws upon intact left-hemisphere structures that have evolved to carry
out phonological and syntactic analyses. Language functions carried out
277
with the right hemisphere depend excessively on semantic and pragmatic
factors; this approach to language proves adequate for most mundane
communication but unsatisfactory for the appreciation of subtle grammat-
ical distinctions.
Even as certain regions of the brain seem preordained to carry out
certain kinds of functions, other considerations militate against a purely
"plastic" perspective. Work in experimental psychology documents that
organisms are "prepared" to master certain behaviors and "counterpre-
pared" to learn other ones. Thus, for example, rats can quickly learn to run
or jump to escape shock but with only the greatest difficulty to press a lever
in order to effect the same escape. Similarly, jumping to avoid a shock
seems a natural or "prepared" response; but if a rat must jump in a box
with a closed lid, learning will be slow and uncertain. It seems reasonable
to contend that normal children are "prepared" to learn language very
quickly, and that at least some children are "prepared" to master the tonal
system of their culture's music with little exposure: it would not be sur-
prising to find children having great difficulty in learning some kind of
"nonnatural" language or musical system contrived by a diabolical experi-
menter or dictated by an omnipotent monarch.
The Neural Base of Cognition: Studies of Two Systems
Great scientific debates rarely die out entirely, particularly when there is
a core of validity in the rival perspectives (Holton 1984). Such may be the
case with the controversy between the localizers and holists, which has
been evident since the time of Descartes and highly vocal since the period
of Gall. Studies conducted over the last twenty-five years, in the wake of
the Hixon Symposium, have, however, helped to frame the debate. It is
now conceded that, at least at the level of sensory processing, the nervous
system is specifically constructed to respond to certain kinds of informa-
tion in certain kinds of ways. There is also evidence for "neural commit-
ment" at much more molar levels of representation, even extending to the
two cerebral hemispheres. To this extent, Lashley greatly overstated his
case. On the other hand, impressive evidence continues to accumulate
documenting the resilience and plasticity in the nervous system, particu-
larly during the early phases of development. At such times, even organ-
isms deprived of the usual neuro-anatomical structures are able to adapt
and to carry out requisite functions, sometimes without incurring exces-
sive costs. In this regard, the view of the holists and the mass-action
278
proponents remains tenable. As a tentative conclusion, then, it seems that
some localization is accepted by all, but that important islands of plasticity
remain within this general framework.
But while great debates are seldom silenced altogether, they can be-
come muted. Much less energy is devoted nowadays to debates about
localization as against holism (or specificity as against plasticity), at least
in terms of the operation of the nervous system. Instead, studies have
reverted to a more circumscribed terrain. Neuroscientists are devoting the
bulk of their time to the careful study of specific systems in specific
organisms; they are guided by the hope that these systems can be well
understood in their own terms, and that the knowledge obtained thereby
might ultimately inform more general discussions of the neural basis of
cognition, including the controversial issue of reductionism. From the
many examples that could be cited, I shall mention two; these are taken
deliberately from very different levels of analysis and may contribute in
diverse ways to an emerging cognitive science.
Eric Kandel Bridges a Gap

Recently Eric Kandel and his associates at Columbia University have
succeeded in bridging what had once seemed a vast gap-that between the
functioning of the individual nerve cell and the behavior of organisms
(Kandel1979). This team of researchers has made this leap by focusing on
a very simple organism-the snail Aplysia californica-whose nervous sys-
tem can be readily described and which is also capable of simple forms of
learning, such as habituation, sensitization, and classical conditioning. By
studying these processes in the snail, Kandel and his associates have shown
that these elementary aspects of learning are not diffusely distributed in
the brain but rather can be localized in the activity of specific parts of
neuronal networks. In fact, some learned behavior in the snail may involve
as few as fifty neurons.
In Kandel's account, learning results from an alteration in the synaptic
connections between cells; rather than necessarily entailing new synaptic
connections, learning and memory customarily come about as a conse-
quence of alteration in the relative strength of already existing contacts.
In fact, Kandel and his colleagues have shown that the amount of chemical
transmitter released at the terminals of neurons is crucial in altering synap-
tic strength. Thus, for example, when the tail of a sea slug is shocked, a
neurotransmitter is released. An alteration occurs in the pores of the neu-
ron, so that more transmitter is released in response to a later impulse.
Then the next time the tail is shocked, the neuron quickly "remembers"
to send out chemical commands to retract the siphon.
279
Kandel summarizes his position on the relationship between innate

and experimental inputs to such learning:
The potentialities for many behaviors of which an organism is capable are

built into the basic scaffolding of the brain and are to that extent under genetic and
developmental control. Environmental factors and learning bring out these latent
capabilities by altering the effectiveness of the pre-existing pathways thereby
leading to the expression of new patterns of behavior. (1982, p. 35)
Thanks to this work, we can now glimpse-perhaps for the first time-
what learning entails at the chemical and neural level. The unsettling
question to be addressed by cognitive science is whether, as other more
complex forms of behavior are similarly described, there will remain a felt
need for a separate explanation at the representational level.
The Song of Birds

A dramatically different, but equally suggestive line of study comes
from work on the songs of birds by Fernando Nottebohm and his col-
leagues at the Rockefeller University (Nottebohm 1970, 1980; see also
Konishi 1969; Marler and Peters 1977). Many species of birds sing songs,
and it is well known that these species, and sometimes even subspecies,
can be recognized by their peculiar song. But where does this song come
from, and how does it acquire its characteristic pitches, tempo, and lilt?
It turns out that the answer differs from one species to another. In
some cases, the species song is part of the bird's birthright. Thus, in the
ringdove, a certain song is sung by every male member of the species. No
feedback or external stimulation is needed. In fact, even if the bird is
deafened at birth, it will still sing its species song.
There is less of a lockstep route toward the acquisition of song in most
other species. Typically, birds begin with a period of babblelike subsong,
followed by a period of plastic song, where syllables are repeated or re-
hearsed until they constitute short phrases. Finally, within a year, the
plastic song gives rise to a stereotyped song similar to the songs produced
by other normal adult males.
From the neuroscientific perspective, it is instructive that various dep-
rivations exert predictable influences on the course of song development.
Canaries, for example, require auditory feedback for normal development.
They can, however, go on to produce a well-structured song even in the
absence of hearing the vocalizations of other members of their species:
their own songs suffice. In the chaffinch, however, both auditory feedback
of one's own song and exposure to the songs of other birds are needed if
the chaffinch is to produce a full normal song. If deafened within the first
280
Neuroscience: The FlirtaHon with ReducHonism
three months of life, the chaffinch will produce an extremely abnormal
song, which may prove to be little more than a continuous screech. How-
ever, should it be deafened after it has learned its full song, there is no
discernible deterioration in performance.
Bird song is one of the few instances of brain lateralization among
infrahuman animals. Just as the left hemisphere of the brain of humans is
critical for linguistic competence, so the left hypoglossal nerve in the bird
proves crucial for its production of song. One can produce aphasia (or
amusia) in a bird by destroying the left portion of its nervous system. But
the aphasic canary can recover its prior songs because the homologous
pathways of the right hemisphere have the potential of being exploited.
In this recovery of function, songbirds are more fortunate (because their
nervous systems are more plastic) than adult humans.
The work of investigators like Nottebohm and Kandel is based on the
premise that much can be learned at this point through the careful study
of a single system in a single organism. As Peter Marler, another pioneering
researcher on bird song, has remarked: "The research on birdsong learning
is slow and laborious. But I think in the long run this may be the only way
in which we will ever gain proper understanding of the issue that confronts
us ... namely [in this case] the genesis of natural categories in the percep-
tion of animals and man" (1982, p. 93).
To be sure, the two research efforts proceed on somewhat different
assumptions. Kandel hopes that by studying habituation and conditioning,
he will eventually illuminate processes known to occur in a wide range of
organisms, including humans: the assumption here is that certain learning
processes, sometimes modified as "horizontal," cut across all manner of
content (from learning music to mastering drawing), and that these can be
found in relatively analogous fashion across diverse organisms (from ap-
lysia to man). The Nottebohm line of research, on the other hand, proceeds
along a different vein. Bird song is a behavior that clearly exists only in
birds, though it may conceivably have some phylogenetic ties to human
music or to human language. It is, in any case, a self-contained system,
which allows various kinds of experimental manipulation. Interest in bird
song is consonant with a belief that much cognitive activity is "vertically
organized": that there exists a domain called "song" which may well
follow rules different from other domains and is best understood in its own
terms. Any generalizations that may be validly extended from bird song
to other systems in other organisms-or even from one bird species to
another-will only emerge after careful study of these systems on their
own terms. In neuroscience, no less than in psychology and artificial intel-
ligence, a tension can be discerned between a "modular" and a "general
problem-solving" or "central-processing" point of view.
281
II I THE CoGNITivE SciENCES : HISTORICAL PERSPECTIVE
The remarkable success of the work of Hubel and Wiesel and their
collaborators has given fresh impetus to the belief that systems may work
in their own way. In 1978,· David Hubel put it this way:
We are led to expect each region of the central nervous system has its own
special problems that require different solutions. In vision we are concerned with
contours and direction and depth. With the auditory system, on the other hand,
we can anticipate a galaxy of problems relating to temporal interactions of sounds
of different frequencies and it is difficult to imagine the same neural apparatus
dealing with all of these phenomena .... For the major aspects of the brain's
operation no master solution is likely. (Hubel 1978, p. 28)
Such sentiments, articulated by one of the masters of contemporary neuro-

physiology, tilt in favor of attention to the specific properties of specific
sensory detectors and challenge those in search of a single mode of neural
explanation which pertains equally to different forms of behavior.
The work of Kandel (and to a lesser extent of Nottebohm) raises
afresh the issue of reductionism. It seems to some observers that an ac-
count of the classical psychological phenomenon of habituation in terms
of neurochemical reactions is an important step on the road to the absorp-
tion of cognition by the neurosciences. Once the basic mechanisms of
learning have been described in this way, no additional level of explana-
tion will be needed; in a way that would please such behaviorally oriented
philosophers as Richard Rorty, these reductionists believe there is really
nothing more to be said when neurophysiology has had its say. Yet most
scientists of a cognitive persuasion feel that such accounts, while informa-
tive, will still prove tangential to their ultimate interests. As psychologists
John Marshall and John Morton declare:
The relationship between learning theory and natural behavior is only to be

determined through functional representations of what the organism's nervous
system does, not what it is. With simple organisms such as Aplysia this relationship
can most readily be established in terms of its neurophysiology and neurochemis-
try.... With humans it can at best only be done abstractly .... Suppose it turned
out that all human synapses were equivalent to Aplysia s and suppose that all the
behaviors of such a synapse were expressible in terms of learning theory. Suppose
further that we had all the human neurobiological information there was to have.
We might then have an account of natural human behavior ... but we would not
have an explanation in terms of the questions we really wanted to ask. (Quoted
in Fox 1983, p. 1222)
Pribram 's Holographic Hypothesis

With this heightened interest in the operation of specific neural and
behavioral systems, one might think that proponents of a more holistic
282
view of the brain had been stilled. Not so. Neuroscientist Karl Pribram
argued, in his 1971 book Languages of the Brain, that a belief in specific feature
detectors a Ia Hubel and Wiesel can take one only a limited way: in his
view, the brain is better analogized to a holographic process. Holography
is a system of photography in which a three-dimensional image of an
object can be reproduced (with the appearance of the third dimension
preserved) by means of light-wave patterns recorded on a photographic
plate or film. A hologram is the plate or film with the recorded pattern:
information about any point in the original image is distributed through-
out the hologram, thus making it resistant to damage. Since waves from
all parts of the object are recorded on all parts of the hologram, any part
of the hologram (however small) can be used to reproduce the entire image.
A hologram can store a great deal of information in a small space; indeed
some ten million bits of information have been usefully stored holograph-
ically in a cubic centimeter.
According to Pribram's holographic view, all parts of the brain are
capable of participating in all forms of representation, though admittedly
certain regions play a more important role in some functions, and other
regions are more dominant for other functions. In his view, just as many
holograms can be superimposed upon one another, so can infinite images
be stacked inside our brains. Perhaps when we recall something special, we
use a specific reconstruction beam to zoom in on a particular encoded
memory. Pribram also fixes upon another quality of the hologram: the fact
that it records the same wave front over its surface, repeating it over and
over. Even if only some of a shattered hologram is left, it will still suffice
to reconstruct the entire image (Hooper 1982).
While the holographic analogy has engendered considerable skepti-
cism, many neuroscientists remain sympathetic with Pribram's goal of
showing that the nervous system is not simply a collection of specific
modes of processing; they cling to the possibility that important forms of
knowledge remain, as Lashley had so fervently believed, widely dispersed
throughout the brain. Pribram notes that 11the properties of holograms are
so similar to the elusive properties that Lashley sought in brain tissue to
explain perceptual imaging and engram encoding that the holographic
process must be seriously considered as an explanatory device" (1982, p.
176). Moreover, in a manner reminiscent of the view of limited plasticity,
Pribram himself has recently begun to speak of a 11limited holograph,"
hoping thereby to avoid some of the pitfalls of a full-blown holographic
account. Eric Harth comments:
What interests brain theorists about the hologram is this quality of a distributed
memory (the phrase is Lashley's): Every piece of the hologram says a little bit about
283
II I THE CoGNITivE SciENCEs : HISTORICAL PERSPECTIVE
every part of the scene, but no piece is essential. The other intriguing fact is that
one can superimpose any number of holograms on the same piece of film, and then
reproduce the images of the original scenes one by one without interference from
the others. (1982, p. 88)
The analogy with Lashley's view of the neocortex is striking and may help
explain the continued appeal of a holographic theory of memory.
Three Historical Moments

We might single out three moments in the age-old debate about the
degree of localization of representation in the human nervous system. The
first moment involved scientific hunches. When Descartes located the soul
in the pineal gland, when Gall spoke about the representation of amative-
ness and of criminality in different lobes of the brain, each was announcing
claims without benefit of experimental evidence.
A significant step forward took place when it was possible to examine
the effects of injuries to discrete areas of the nervous system. When Fritsch
and Hitzig lesioned specific sites in the nervous system of dogs, when
Broca and Wernicke looked at the effects of strokes in the human cerebral
cortex, they were able to substantiate correlations between regions of the
brain and forms of behavior.
Finally, when Hubel and Wiesel recorded from discrete cells in the
visual cortex of the cat, it became possible to ascertain with great specifi-
city the function of particular units and the circumstances under which
they would function (or fail to function) normally. So powerful and precise
was this technique that most researchers stopped studying the effects of
damage to large and undemarcated regions of the brain, except in human
beings who had the misfortune to suffer injury to portions of their brains.
It is in terms of these three historically stacked events in neuroscience
that current debates about localization and plasticity can now be framed.
In today's climate, it is possible to utilize highly sophisticated radiological,
electrophysiological, and chemical assaying techniques to study the struc-
ture and the functions of the nervous system, the processes whereby it
develops, the effects that follow upon various kinds of pathology. And the
thrust of these various lines of research has had two effects: first, to render
localization as a more plausible general orientation; second, to direct the
attention of active scientists to the operation of specific systems, rather
than to continued debate on broad conceptual issues.
And yet it is far too early to claim that the pendulum has stopped
swinging, or that the pivotal questions motivating neuroscience have been
answered. As Pribram's interest in the holographic aspects of the nervous
284
system indicates, and as the recent studies of recovery of function in young
animals make clear, the voices in favor of mass action and plasticity have
not been stilled. Moreover, within the specific areas of higher cognitive
functioning (as we shall see), there remain debates about even the most
basic issues. Neuroscientists still do not agree about just how to describe
the functions of the particular cells that Hubel and Wiesel have studied
(for example, in terms of sensitivity to spatial frequency rather than the
detection of line orientation) (Pribram 1980, p. 58); nor how to characterize
the various difficulties that aphasic patients demonstrate; nor how best to
label the functions of the right hemisphere in a normal individual, one who
suffers from unilateral brain injury, or one whose cerebral hemispheres
have been separated for therapeutic reasons. Moreover, even if localization
seems (on the whole) to be more tenable than holism, it is now apparent
that reductionism is a separate issue.
Will Neuroscience Devour Cognitive Science?
In some ways the neurosciences are different from the other cognitive
disciplines (even as philosophy, the wholly nonexperimental discipline, is
also different). Researchers in the neurosciences stand out from their cog-
nitive-scientific peers because they most closely partake of the model of
the "successful" sciences of physics and biology, because they can most
readily state their questions unambiguously and monitor progress toward
their solutions. While defenders can be found for the propositions that
psychologists or anthropologists have made little progress, or have failed
to define their central issues with sufficient progress, few, if any, informed
observers would level the same charge at the neurosciences. It is for this
reason, in part, that I have avoided an excessively historical approach in
this chapter, and focused instead on one central issue in the field-in order
to show how the question has been framed, to look at the various altera-
tions in perspective, and to evaluate neuroscientific conclusions after a
century of work on the question.
Yet it should be clear that while the neurosciences are different (and
who would decry that difference?), they are not all that different. As I have
observed, fundamental debate continues on many of the central questions.
The uncertainty I have documented reflects the fact that, while rapidly
growing, neuroscience is a young field and is still in the process of d~fining,
rather than resolving, many principal issues. And the uncertainty points
to another consideration as well.
285
There are those (perhaps a majority) within the neurosciences who

would maintain that cognitive scientific concepts and concerns are not
relevant to a biologically oriented science. On this reductionist view, the
goal of any natural science should be to explain phenomena at the most
elementary level possible. Just as physics moves toward explanation at the
subatomic level, and biology searches for explanation at the genetic and
molecular levels, so the neurosciences should become oriented increasingly
toward the nerve cell and the chemical and electrical events that occur
within. So long as the science remains relatively immature, it may be
necessary, for the time being, to carry out psychological experiments or to
engage in computer simulations of behavior. But once the appropriate
neuroscientific studies have been carried out, explanations that feature
behaviors, thoughts, actions, schemas, or other molar or representational
concepts should become superfluous.
There is a parallel reductionist tendency within the neurosciences
themselves. So long as the science remains immature, it may be necessary
to speak about occurrences at molar neural levels (for example, within a
column of cells, a region of the occipital lobe, or even an entire lobe or
hemisphere of the brain). But ultimately such talk should disappear, as the
same occurrences can be accounted for at the level of a particular cell. Thus,
a strong reductionist position proceeds in two directions: it jettisons the
terminology of psychology or phenomenology as rapidly as possible; and
it moves from grosser regions of the nervous system to more specific sites.
There are, of course, weaker reductionist positions, which permit some
cognitive talk or some discussion of more molar neural systems; but they
share with strong reductionism a skepticism about any concept whose
relation to well-established neural facts remains obscure.
From the perspective of cognitive science-one that I happily adopt
here-these lines of argument are untenable. While cognitive scientists
may differ in the extent to which they are interested in, or knowledgeable
about, new discoveries and favorite models of brain scientists, they are
agreed about one crucial matter: that is, one cannot have an adequate
theory about anything the brain does unless one also has an adequate
theory about that activity itself (Mehler, Morton, and Jusczyk 1984). It is
not possible to study perception-even its most fine-grained forms-with-
out a theory of perception. It is not possible to study classification without
having a theory of categorization, without substantive knowledge about
domains being categorized, as well as understanding the philosophical
issues involved in constructing or deploying a category. It is ill-advised to
talk about mind, self, action without considerable familiarity with the
pitfalls that have long resided as alligators in these particular mentalistic
swamps. To paraphrase Wittgenstein, one can know every brain connec-
286
tion involved in concept formation, but that won't help one bit in under-
standing what a concept is.
From this perspective, it is not possible to enter into the nervous
system as a disinterested observer who is simply chronicling the facts (as
many neuroscientists assume they are doing). Both the topics studied, and
the ways in which they are studied, will reflect implicit theories: theories
about what perception, cognition, or language are; what is important in
each; and how each of these processes occurs. Risks of generalizing from
lower animals to human beings are severe; and, in any event, the less
self-conscious one is about procedures, the more likely it is that one will
make naive errors.
Thus, to take an example from language, a neurologist ignorant of
linguistics might rely on naive intuitions about language: one would there-
fore describe an aphasic patient as unable to use "small words" or to
"speak in full sentences." But a linguistically trained observer will im-
mediately be able to pose questions and introduce distinctions at a subtler
level: Which grammatical categories pose trouble? Are these troubles ap-
parent across different linguistic contexts? Do they correlate with other
phonological, syntactic, lexical, or pragmatic difficulties or distinctions?
Having recourse to such linguistic insights does not guarantee an accurate
analysis. (Sometimes, indeed, too strong an investment in a particular
linguistic theory can blind the observer to counterparadigmatic phe-
nomena.) Yet, to avoid any knowledge of linguistics is equivalent to blind-
ness-by-choice: and so it is now recognized that any aphasia research team
should include psychologists and linguists as well as neuroscientists.
For these reasons, I think it best to regard neuroscience as one of the
border disciplines of cognitive science. Just as anthropology represents a
kind of upper bound for the investigation of cognitive phenomena, so the
neurosciences represent a kind of lower bound. Many of the phenomena
investigated by neuroscientists are either accounted for perfectly ade-
quately without any reference to the representational level or allow but a
subsidiary role for representational aspects: thus, an account of habitua-
tion in Aplysia does not cry out for representational analysis. Such
phenomena do not belong to the mainstream of cognitive science, any
more than do discussions of the religious system of an aboriginal tribe. But
once neuroscientists begin to invade domains that entail more complex
forms of mentation-for example, the domains of language or perception
of objects or logical problem solving-there is no longer any possibility of
finessing these representational issues. It is at this point that interdiscipli-
nary cooperation becomes an imperative. And it is at this point that the
cognitive challenge arises: how best to build explanatory bridges between
the level of the neuron and the level of the rule or the concept.
287
Thus, much as they would like to, neuroscientists cannot afford to

isolate themselves from others in the cognitive sciences. While fields like
psychology, philosophy, linguistics, or anthropology may not have the
answer to questions about perception, memory, learning, or language,
work in the fields over the decades is relevant to any contemporary discus-
sion. Ignorance of them ill serves even the most gifted neuroscientist. Even
those who much prefer a "wet" approach (where one "opens up" the brain
and looks at, or listens to, what is there) to a "dry" approach (where one
tries to develop an adequate theory of the behavior observed) are increas-
ingly recognizing that the solution to their problems is more likely to come
about through extended interdisciplinary collaboration. I shall examine
directly several such promising interdisciplinary efforts in part III.
288
PART III
TOWARD AN
INTEGRATED
COGNITIVE
SCIENCE: PRESENT
EFFORTS, FUTURE
PROSPECTS
Introduction
In part II, I reviewed highlights from the history of six separate disciplines,
which collectively constitute the cognitive sciences of today and may some-
day be integrated into a single cognitive science. While I have considered
each discipline separately-to monitor how each discipline developed in
terms of its internal standards, the principal issues it confronts, and its
workaday methods-no scholarly discipline operates in a vacuum. There
have been clear conversations among various disciplines-for example,
between philosophy and psychology, between linguistics and anthropol-
ogy, even between artificial intelligence and neuroscience. To put it in the
terms of psychologist E. G. Boring (1950), one can discern the Zeitgeist at
work; or, to use the term of the structuralist historian Michel Foucault
(1970), a common episfeme has been imposed on disparate disciplines.
To convey the factors that seem to have been at work across disparate
disciplines, I shall briefly sketch the history of cognitive science as if it
were a single coherent field. Our prototypical science goes back to the
Greeks-to the writings of Plato and Aristotle, the topics raised in the
Meno. It is here that questions about the nature, the status, the sources, and
the use of knowledge were first raised. The agenda of cognitive science was
fleshed out during the philosophical flowering of the seventeenth and
eighteenth centuries-in the debates between rationalists (like Descartes)
and empiricists (like Locke and Hume) and in the attempts at synthesis put
forth by Kant around 1780 and by Whitehead and Russell around 1910.
Here we see the first intimations of the topics that exercise the cognitive
scientist of our time.
In the nineteenth century, the province of our prototypical science
became that of empirical rather than of introspective scientists. The
unquestioned successes of physics and chemistry, and the inspiring exam-
ple of Darwin in the biological sciences, fanned hopes for a comparable
291
I I I I TowARD AN INTEGRATED CoGNITIVE SciENCE
science of human behavior and thought. Aided by the advent of graduate

studies, the rapid growth of technology, and the emergence of a society at
once affluent and yet beset by new social and economic problems, the
disciplines of psychology, linguistics, anthropology, and neuroscience all
began to emerge within a few decades of one another. While the focus of
these disciplines differed, the forerunner of our prototypical cognitive
science was based on the premise that human behavior and thought have
evolved over the millennia, have adapted to the various environments in
which cultures arose, and could be subjected to study by empirical and
perhaps experimental methods. It should be possible at last to secure
reliable answers to age-old questions about human perception, language,
classification, and reason.
The leaders of these new disciplines at the end of the nineteenth
century were ideally suited to the task: they were not only deeply rooted
in the natural science and philosophy of the era but were also imaginative
empiricists and inveterate organizers, capable of launching societies, edit-
ing journals, and arguing tirelessly with sympathizers and skeptics. Helm-
holtz, James, Wundt, de Saussure, Frege, Tylor, and Boas are just some of
these pioneers.
These researchers succeeded in establishing new sciences, but their
specific initial agenda-grandiose and optimistic-did not always take
hold. There arose in the early part of the twentieth century a series of
reactions to the visionary first generation. For the most part, this reaction
was atomistic and functionally oriented: behavior was to be analyzed in
terms of its constituent elements and viewed in light of its role within an
organism's life and with respect to the goals of a society. The behaviorist
vogue swept over psychology (Watson) and linguistics (Bloomfield); func-
tionalism dominated anthropology (Malinowski); logical empiricism held
sway in philosophy (courtesy of the Vienna circle). The only exception to
this ideal pattern was neuroscience, where (perhaps appropriate for a "bor-
der discipline") the reaction was in the opposite direction: the localization
views of an earlier generation were swept aside for a while by a holistic
perspective.
The several disciplines were not completely enveloped by the behavi-
orist mentality. Within each, some scholars adapted a holistic or a structur-
alist position. The Gestalt psychologists and Piaget, the Prague linguists,
the structural anthropologists like Radcliffe-Brown, kept alive the broader
visions of a science of mind. And during the very time that mentalism was
in disrepute, mathematicians and engineers were laying the groundwork
for the breakthroughs that would eventually undermine the hegemony of
behaviorism.
Behaviorism prevailed in the United States and, to a lesser extent,
292
Introduction
abroad in the period from 1920 to 1950. But as the Hixon Symposium, and
the Macy conferences were being held, as Craik and Turing and Wiener
wrote, and as Bruner, Chomsky, Levi-Strauss, Miller, Newell, and Simon
were pursuing their studies, a cognitivist revolution was brewing. Scholars
were discovering (or rediscovering) the centrality of high-level linguistic
and conceptual activities and the utility of the new tools of computing for
investigating these phenomena. What were once iconoclastic statements
by such scholars became (with surprising rapidity) the new orthodoxy.
Behaviorism was not so much defeated as rendered irrelevant by frontal
approaches to human cognitive processes.
In the late 1950s and early 1960s, there was excitement in the already
established disciplines and euphoria in the new, quintessentially cognitive
discipline of artificial intelligence. But the promise of the cognitive sciences
would not be so readily realized. There ensued serious debates about how
best to approach the discipline: whether by top-down or bottom-up tech-
niques; by general problem solving or expert systems; by programmatic
long-term experimental work, highly selected demonstrations, or detailed
case studies. And when it became clear that early predictions would not
be confirmed, a reaction set in, during the early 1970s: perhaps a cognitive
science was not to be, at least during the lifetime of its most fervent
enthusiasts.
There is another, more attractive possibility, however. Perhaps the
individual cognitive sciences have gone as far as they can within each of
their disciplinary constraints and paradigms. What were once merely polite
and brief conversations among them need to be converted into full-scale
cooperative research efforts on problems central to several of them. Stem-
ming both from a realization that many scientific problems are simply too
complex to be handled by a single discipline, and from a genuine attraction
to the methods and concepts worked out in neighboring disciplines, a
growing number of scientists-many of them raised in the post-1950s
environment-are switching allegiance from single disciplines to the
broader practice of cognitive science and are engaging primarily in inter-
disciplinary pursuits.
Accordingly, in the final chapters of this book, I review several re-
search efforts that qualify for the label cognifive-sdentific. These represent
several disciplines-either combined in one person or laboratory, or in-
volving collaboration across individuals or laboratories-and are intended
to answer the long-standing philosophical questions that originally ener-
gized thinkers in classical times. Of course, many different examples could
have been chosen, and my selection is not meant in any sense to be
definitive. I chose lines of research that involve the central questions in
cognitive science and that are considered by the cognitive-scientific com-
293
II I I TOWARD AN INTEGRATED COGNITIVE SCIENCE
munity to be of high quality. Not beyond criticism (no good scientific work
is immune to critical analysis), this work has rather generated sophisticated
criticism and suggested approaches that may eventually resolve such criti-
cism. Indeed, in what follows, I have not hesitated to cite significant
criticism and, where appropriate, to state my own reservations. I have
chosen to focus on four sets of question that have generated cognitive
science of quality.
In chapter 10, I review contemporary efforts in artificial intelligence,
aided by neuroscience and psychology, to explain how humans perceive
forms and objects. Central to this question is the pathbreaking work of
David Marr and his associates, as well as criticisms put forth by}.}. Gibson
and other believers in "direct perception."
In chapter 11, I consider the status of visual imagery: What is meant
by talk of imagery, and is it proper to treat images as a means whereby
knowledge is represented? My survey here focuses on the work of Stephen
Kosslyn, in psychology and artificial intelligence, and on various philo-
sophical and computational objections to this work put forth by Zenon
Pylyshyn.
In chapter 12, I tum to the issue of how human beings classify objects
and elements in their world. A point of departure are studies of color
naming carried out by the cross-cultural psychologist Eleanor Rosch and
two linguistically oriented anthropologists Brent Berlin and Paul Kay.
Work here impinges on the philosophical issue of whether human classifi-
cations are arbitrary or motivated, as well as on the relationship between
the languages we use and the ways in which we think.
Human rationality is the subject of chapter 13. While philosophers
since Greek times have pondered the nature and the extent of human
rationality, recent work by researchers like Amos Tversky and Philip John-
son-Laird severely questions the model of man as a logical thinker. This
critique, drawn from psychology, linguistics, and artificial intelligence,
has engendered considerable philosophical debate-reason discoursing on
irrationality.
A survey of the best work, and the sharpest critiques of that work,
is one way to evaluate current cognitive science. But, in the end, the way
that the field presents itself-its overall charter-provides an equally im-
portant test. For the most part, I have simply presented this charter and
allowed the field to speak for itself through its history and its work. In the
end, however, I suspend this authorial pose and present my own conclu-
sions about the movement whose early gropings stimulated me to write
this book.
294
10
Perceiving the World
How can we recognize a circle as a circle, whether it is large
or small, near or far; whether, in fact it is in a plane perpen-
dicular to a line from the eye meeting it in the middle and
is seen as a circle, or has some other orientation, and is seen
as an ellipse? How do we see faces and animals and maps
in clouds, or in the blots of a Rorschach test?
-NoRBERT WIENER
Perennial Puzzles of Perception
When philosophical reflection begins-whether in a child or a society-it

typically focuses on the perception of the external world. The naive indi-
vidual is struck by the existence of a world of objects which one can see,
so long as eyes are open and there is sufficient light. The objects move, we
may also move, and yet we continue to see a stable, organized world. We
can also see forms and colors, possibly only when they are perceived as
part of an object, but perhaps also prior to, or instead of, the objects
themselves. And most of us believe that we can also perceive in the ab-
sence of objects by relying on our memories, our mental images, our
powers of imagination.
These are seemingly straightforward observations, but the history of
philosophy and psychology testifies to the difficulty of finding acceptable
accounts of how ordinary percepts come about. Problems emerge almost
immediately. Does all the information necessary for accurate perception
exist in the external world, or do we bring expectations and knowledge
to the perceptual encounter? The world is presumably possessed of a
295
third dimension, but how does this third dimension get recorded on the
retina and then reconstructed within one's head? How different is it to
perceive the world as compared with a picture of the world or with an
image of the world within one's head? Why do optical illusions persist
even after they have been recognized as such? And how do various vi-
sual impressions and images relate to thinking? Are they thoughts in
themselves, do they provide the vehicles of thought, do they reflect the
manipulation of symbolic entities, or are they but epiphenomena, ves-
tiges that do not materially contribute to our ability to know, to learn, or
to understand 7
Questions like these have been pondered ever since philosophy began.
Indeed, the Greeks modeled knowledge upon vision; they invested consid-
erable effort in understanding how we come to know the visible world and
how this knowledge may contribute to--or constitute-general under-
standing. The pre-Socratics were already engaged in debate: Metrodorus
of Chios counseled his followers to disregard evidence from one's senses
and to pay attention to belief; Democritus in turn acknowledged that all
knowledge rests on perception (Barnes 1979). Plato believed that the soul
makes perception possible, while Aristotle was more interested in discov-
ering how the eye actually works. In more recent times, rationalists and
empiricists revisited these issues. Whereas Descartes minimized the impor-
tance of sensory organs, the empiricists saw them as the point of origin for
all knowledge.
Before long, practicing scientists joined these debates. In fact, proba-
bly a majority of psychologists have begun with an interest in perception,
and there has been more unambiguous progress in understanding percep-
tion than in illuminating other mental processes. In my review of psychol-
ogy, I have already touched upon some dominant themes: Hermann von
Helmholtz's belief that there is insufficiently accurate information in the
stimulus itself, and that much of perception accordingly depends upon
unconscious inferences about the scenes that have been observed; J.J.
Gibson's rejoinder that the senses can directly pick up from the environ-
ment information that is needed for survival; the sensationalist's conten-
tion that perception begins with the detection of elementary bits of sensa-
tion, out of which it builds up to more complex objects and forms; the
Gestaltist's contrary assertion that one first perceives overall form, in a
top-down fashion.
Two new lines of research entered discussions about perception. One
line, already reviewed in chapter 9, sought to determine the sensitivities
of particular nerve cells. If the functioning of all neurons and sets of
neurons were to become known, there might be no need for more abstract
or higher-level descriptions of perceptual processing.
296
Computer Simulations
With the advent of computers, another group of scientists became con-

cerned with the processes of visual perception. For them, the initial chal-
lenge was less to figure out just how perception occurs in the human being
than to understand how perception can be possible in any kind of organism
or mechanical device. They approached this engineering challenge by attempt-
ing to design "seeing" machines and programs: provided with some visu-
ally presented information, these mechanisms could so parse that informa-
tion as to determine the nature and form of presented objects, patterns,
or scenes.
After false starts in the late 1950s and early 1960s, students of percep-
tion initiated a line of work called scene analysis (Boden 1977; MacArthur
1982). In a typical research program, a picture (usually a line drawing)
serves as input to the computer, and the computer's task is to interpret the
picture in terms of the objects depicted in it, stating the relations of objects
to one another and describing the objects themselves. It seemed evident
that the "real" world was too complex for current technology to parse, and
the most effective programs were devised to work with artificial mi-
croworlds (of the sort that Terry Winograd had used for his language
understanding program SHRDLU). As we have seen (page 170), research-
ers like David Waltz were able to devise programs that could unambigu-
ously interpret most line drawings of a scene containing blocks. These
approaches of the early 1970s were, however, limited in various ways. For
one thing, the programs tended to use artificial line drawings, rather than
real objects or photos of real objects. Moreover, the scenes usually con-
tained a highly constrained set of forms drawn from a particular artificial
microworld. Computation at the level of the image itself was minimized:
there was emphasis on the manipulation of symbolic descriptions of a
scene, the kinds of operations in which computers excel, as well as ample
reliance on previous knowledge about forms of this sort. In contrast, the
rich information available, under normal conditions, to an ordinary viewer
was not taken into account. More generally, even when a scene was parsed
successfully, there was a feeling that the program used clever engineering
tricks, rather than psychologically plausible mechanisms.
If further advances were to take place in computer vision, it was
necessary to consider not just how to perceive a particular kind of picto-
rial image or microworld, but rather to attack the process of perception at
a fundamental level; the aim would be to come up with mechanisms
that, whether or not similar to those used by animate perceivers, could at
297
I I I I TowARD AN INTEGRATED COGNITIVE SciENCE
least handle the whole range of perceptible scenes and the gamut of per-
ceptual tasks-including perception of motion, depth, surface textures,
and other variables that make human perception so powerful. The com-
bination of insights now available from perceptual psychology, neuro-
science, and artificial intelligence made it possible, probably for the first
time, to put forth a reasonably complete account of the early phases of
visual perception-an account that explained how any organism is able to
perceive a world of shapes and objects. The relevant input to this task
came from many quarters, but the pathbreaking conceptions came largely
from David Marr.
The Work of David Marr
Though tragically short, Marr's research life was amazingly productive.

Trained in neurophysiology at Cambridge, England, he began his scholarly
work with a study of the functioning of the human cerebellum (Marr
1969). He came to realize, however, that the most important questions
about behavior could never be answered simply by looking at the brain
alone-not even if one knew the function of every cell and every connec-
tion; as he said, it was like trying to understand flight simply by examining
the feather of a bird. Just as a complete understanding of flight must entail
an understanding of the constraints preventing any organism or machine
from lifting into flight, and of the factors allowing that organism to over-
come gravity, so, too, a complete understanding of perception can come
about only by means of a theory that considers the actual problems in-
volved in perceiving objects, how these constraints might be counteracted,
and how they may be realized in particular mechanisms, ranging from
computers to brains. This realization stimulated Marr to undertake work
in the artificial intelligence laboratory at Massachusetts Institute of Tech-
nology from 1973 until his death from leukemia, a scant seven years later.
In this brief period, Marr laid out a program of how to approach visual
perception in particular and of how to study knowledge systems in general.
Fundamental to his perspective was the belief that vision is the construc-
tion of efficient symbolic descriptions of the images encountered in the
world. As he once expressed it, the images of the world should yield a
description that is useful to the viewer and not cluttered by irrelevant
information. In opting for a symbolic description, Marr broke decisively
from those researchers who believed in "direct perception," and entered
into the camp of cognitive science.
298
Levels of Scene Analysis
Marr maintained that, to gain such a description, it does not suffice

to have understanding at one level of analysis. One has to be able to
describe the responses of neural cells, to predict the results of psycho-
physical experiments, and to write computer programs that analyze and
interpret visual input in the desired way (Marr 1982). Accordingly, he
proposed three levels of explanation--computational theory, algorithm,
and implementation.
Marr's levels can be introduced through the simple example of a cash
register. To understand such a device, you have to ask, at the most abstract
level, what it does and why. Since its job is arithmetic, you have to master
the theory of addition, which involves understanding the notion of map-
ping pairs of numbers to their sums, and appreciating such principles as
the laws of commutativity-as in 3 + 4 = 4 + 3-and of associativity-
as in (3 + 4) + 5 = 3 +(4 + 5). Such understanding of the theory of
addition constitutes the computational theory of the cash register.
In the world of visual processes, the analogy is the problem of deriv-
ing properties of the world from images of it: in other words, the chal-
lenge of explaining vision in whatever mechanism happens to realize it.
Such a computational theory shows what is being computed and why,
for the purposes of perception, this is a useful piece of information to
compute. In the case of vision, the computational theory would specify
the ways in which a two-dimensional image is related to the three-
dimensional world and how that image can be interpreted-the con-
straints that make it possible to recover the properties of the scene from
the correlative image. Thus, let us suppose that vision includes such pro-
cesses as recovering the shape of a rotating body from fleeting images, or
the capacity for stereoscopic perception (the ability of a binocular mech-
anism to compute depth by combining information obtained from two
slightly different points of view). Any computational theory must ac-
count for how these processes can be carried out given the information
available in the image.
In order for the process actually to run, however, one has to realize
it in some way and therefore chqose some form of representation for each
of the relevant entities. The second level of analysis of a process involves
selecting, first, a representation for the input and the output of the process;
and second, an algorithm (or formal symbol-manipulation procedure) by
which the transformation may actually be accomplished. In other words,
this second algorithmic level specifies how an operation is carried out. In
the case of addition, one might choose Arabic numerals for the representa-
tions; and for the algorithm, one could add the smallest digits first and
299
carry if the sum exceeds 9. Both electrical and mechanical cash registers
follow this type of representation.
In the case of vision, the algorithmic level addresses itself to the
question of the various ways in which a function like stereopsis might
actually be represented and carried out by some mechanism. In this partie~
ular case, Marr and his colleagues had laid out an elegant set of procedures
for computing stereoscopic vision; it turned out, however, that this set bore
no resemblance to the processes apparently used by the brain. Marr
wanted his algorithms to be consistent with what is known about animal
perception: accordingly, that algorithm was replaced by one more consist-
ent with evidence from psychophysics and from neuropsychology.
The third level concerns the device in which the process is to be
realized physically. The same algorithm can be implemented in quite dif-
ferent technologies. Addition can be run on various electrical or mechani-
cal machines but can also be carried out in the brain.
Any task may be achieved by various algorithms, and any algorithm
is susceptible to many realizations in a given hardware. The decision about
which way the algorithm will actually be realized is made by each investi-
gator, and many students of artificial intelligence are unconcerned about
the ways in which human beings happen to accomplish visual perception.
Marr' s interest in the level of realization clearly concerned the possibility
of building computer programs to parse scenes effectively. This was the
essential"existence" proof that an algorithm actually worked. As we have
seen, however, he let his work be guided by the procedures that seemed
to be used by the human brain-possibly because these seemed most likely
to work effectively for inorganic machines as well.
Marr noted that it was easier for scientists to work on the algorithmic
and the mechanical realization level-levels that lend themselves to actual
experimentation. But in his view, the level of computational theory was
most important to tackle at the present time. Here Marr reflects his belief
that the nature of the computations that underlie perception depends more
on the computational problems that have to be solved by any system, than
on the particular hardware in which the solutions happen to be imple-
mented in the most familiar instances.
Marr cautioned against excessive concentration on the nature of brain
processes alone; as he noted, even total knowledge of anatomy and physi-
ology would not allow one to understand why neurons have receptive
fields. To understand how the neurons of the visual system actually ac-
complish their tasks, one must draw upon mathematical principles in-
volved in interpreting images. Yet Marr was conscious that no discipline
in itself could unravel the mysteries of perception. A true cognitive scien-
tist, he once declared:
300
The moral is that ignorance in any of these three fields can be damaging. Just
as the modern physicist has to know some mathematics, so must the modern
psychologist, but the psychologist must also be familiar with computation and
have a clear idea of its abilities, its limitations, the fruitful ways in which to think
about processes and, most importantly, what it takes to understand these processes.
(1982, p. 187)
Two Sketches and a Model

So much for theoretical preliminaries. Having laid out a general ap-
proach to the understanding of vision, Marr and his colleagues devoted
their efforts to specifying steps governing individual processing of visual
images-be it implemented by a machine or by some organism. The goal
of the Marr enterprise was to explain how our brains come to compute
roughly the same symbolic representations starting from initial "gray-
level" variations in illumination. Briefly, Marr and his colleagues described
a series of representations called "sketches" which began with the quickest
and least detailed kind of parsing of visual information and culminated
when the image was seen as a complete collection of objects arranged in
space. (See Marr 1982 and Rosenfield 1984 for fuller accounts.)
The first representation was the primal sketch: this made explicit the
properties of the two-dimensional image, ranging from the kinds of inten-
sity change within a scene (areas of gray, of relative brightness as com-
pared with areas of relative darkness) to a primitive representation of the
local geometry. The next representation, the 2 0 -D slcetch, involves a repre-
sentation from the vantage point of the viewer of the depth and orientation
of the visible surfaces. The final step is the three-D representation. Its
coordinate system is object- rather than viewer-centered; it includes a
representation of volume, the space occupied by an object and not just its
visualized surfaces, as well as an organized array of simple recognizable
shapes of various sizes. In laying out these sketches, Marr and his col-
leagues were outlining the steps through which any mechanism necessarily
passes from the time (or circumstance) in which it first attempts to make
an external scene intelligible, to the time {or circumstance) when the scene
has been apprehended in relatively veridical form.
A closer look at the first steps of visual processing indicates that these
processes are designed to sort out factors of geometry, reflectance of a
surface, illumination of a scene, and determination of viewpoint. The
multiphased procedure of forming the primal sketch involves steps like
detecting intensity changes, representing and analyzing local geometrical
structure, and detecting illuminating effects like light sources, highlights,
and transparencies. The steps reflect the fact that changes in illumination
301
I I I I TowARD AN INTEGRATED CoGNITIVE SciENcE
occur in a scene at the point where edges and changes in surface contour
are likely to occur. This phase ends with a representation that makes
explicit the size and disposition of intensity changes-allowing one to
detect the boundaries in an image and what has caused them. The primal
sketch consists of a set of blobs oriented in various directions; these are
reminiscent of the sorts of features discerned by Hubel and Wiesel's detec-
tors-contrasts, spatial extent, general orientation at a local level. All of
these reductions and simplifications are conceived of as mental representa-
tions or symbolic depictions of the "raw information" transmitted by the
light: perception consists of a series of such simplified sketches en route to
a more veridical view of the world.
Following the achievement of the primal sketch, a number of processes

operate upon it to derive a representation of the geometry of the visible
surfaces. These processes include stereopsis, the use of cues of shading,
texture, occluding contours, and various aspects of motion. In global ste-
reopsis, one attains an internal representation that includes information on
depth, surface orientation, and surface discontinuities. But like the primal
sketch, the 2V2-D sketch is constructed in a viewer-centered coordinate
frame. It depends on a single vantage point and therefore cannot explain
one of the most important facts about visual perception: the perceived
constancy of the shape of an object despite movements on the part of a
viewer.
Marr claims that the goal of early visual processing is to construct a
2Y2 -D sketch. One avoids all the problems of traditional psychological
analysis associated with intuitive distinctions like figure, ground, region,
and object; the various modules of early visual processing and the 2¥2-D
sketch itself deal only with discovering the properties of surfaces in an
image. These occur in exactly the same way, whether one is viewing
persons, animals, trees, or paintings. Only shapes and reflections need to
be made clear to the viewer at this point. As Marr puts it, the 2¥2-D sketch
is the final step before a surface is interpreted (as being a particular object
or set of objects): in fact, it may well mark the end of purely perceptual
processes.
The final step of early visual processing involves the transformation
of shapes from a pure representation that is matched to the processes of
perception into a representation that is suitable for recognition-a set of
meaningful regions. The task now becomes object recognition: needed is
a stable shape description that does not depend on a particular momentary
viewpoint. Thus, the pieces of a shape must be described in terms of a
frame of reference based on the shape itself. A scheme for representing
shape involves the use of a coordinate system and component axes iden-
302
a. b. c.
,..
.........,
..\.:. 'J
.. .
. "
•" '
An image (a), the spatial components of its primal sketch (b), and a
reconstruction of the image from the primal sketch (c) .
•
• ··-·········· ·········· •
• -f·.·.:···t ···-~·-····~·······t r··:::r . . ...
·~.-"II t '-....;-. •
>·~ I" II f ' 0.... r.. •

• -:··...~.·.····~:·:.·. :::·.::.·::':::. ·:.·>-.····~···:~. . . .t- •
..... _._., ' , ..... , ......
·~_ .. 1"11
'
', .......... ,_ •
• ··........... ~ ..... l!....... l ......, .... ·"····o:;.;·····.
• •
A 2¥2-D sketch without a representation of depth information.

Human
rn r~~--~ -f----L_o_w_e_r_A_r....,m
~
Arm
~- Hand
~-1---------1
~
I
'--------i-------'
Organization of shape information in a 3-0 model description. Each

box corresponds to a 3-D model.
SouRCE: From P. H. Winston and R. H. Brown, ArHjida/lnftlligtnct: An MIT Perspective, vol.
2: Undmta..Jing Vrsion, ManipulaHon, ComTJKftr Dtsign, Symbol ManiTJKlalion (Cambridge, Mass.:
MIT Press, 1979). Reproduced by permission.
I I I I ToWARD AN INTEGRATED CoGNITIVE SciENCE
tified from an image which capture what is specific about the objects in
question.
What does this mean in practice? The object is broken down into
components and subcomponents until all of its parts have been uniquely
specified. The model's coordinate system and component axes must be
identified from an image, and the arrangement of the component axes in
that coordinate system need to be specified. Whereas the products of a
primal sketch look like line segments oriented in various directions, the
products of the three-0 sketch look like stick figures composed of pipe
cleaners. This reason is that according to Marr and his colleague Keith
Nishahara (1978), the brain automatically transposes the contours it has
derived from the ZY2-0 sketch onto axes of symmetry that resemble stick
figures. By the time the three-0 sketch has been constructed, the final
result should be a unique description of any object one can distinguish; the
same object should always yield the same unique description no matter
what the angle of viewing; and different representations should reflect
the similarity between different objects, while also preserving whatever
differences may matter.
We have, then, a series of steps through which presumably both
humans and machines must pass in making sense of a scene or an image.
The first computational stage, the formation of the primal sketch, consists
of description of a scene in terms of a vast collection of features like edges,
lines, and blobs-the kinds of feature that may depend upon specific
neural detectors a la Hubel and Wiesel. This primal sketch, an initial
symbolic representation of the image, is formed by processing mechanisms
that are completely independent of any "high level" knowledge about
objects. The second stage involves analysis of the primal sketch by sym-
bolic processes that are capable of grouping lines, points, and blobs to-
gether in various ways. The point here is that generally one can see a round
triangular bulb before knowing that it is a chestnut tree. And, correlatively,
in certain varieties of brain damage, one may be able to see shapes quite
reliably without knowing what objects they represent. Then, in the final
stages, an actual identification of an object, along with its component parts,
is made; and this identification should uniquely determine which object is
being perceived. Top-down knowledge about the nature and construction
of the objects of the world is presumably brought to bear in this last phase
of early visual processing. And so, according to this scheme, the sorts of
knowledge about the world which had earlier appeared essential for per-
ception actually come into play only after shapes have been completely
analyzed.
I have sought to convey the boldness of Marr's approach in artificial
intelligence and, without going into technical details, to suggest the way
304
in which he conceived the steps of visual processing. It is equally important
to convey the enormity of the enterprise. At the level of the primal sketch,
for example, the primitives the algorithms must detect include blobs, ter-
minations and discontinuities, edge segments, virtual lines, boundaries,
groups, curvilinear organization, and other such elements. Marr himself
attempted to describe the process in general terms:
One initially selects roughly similar elements from [the image] and groups
and clusters them together, forming lines, curves, larger blobs, groups, and small
patches to the extent allowed by the inherent structure of the image. By doing this
again and again, one builds up tokens or primitives at each scale that capture the
spatial structure at that scale. Thus if the image was a close-up view of a cat, the
raw primal sketch might yield descriptions mostly at the scale of the eat's hairs.
At the next level the markings on its coat may appear-which may also be detected
directly by intensity changes-and at a yet higher level there is the parallel-stripe
structure of these markings .... at each step the primitives used are qualitatively
similar symbols-edges, bars, blobs, and terminations or discontinuities-but they
refer to increasingly abstract properties of the image. (1982, p. 91)
Implications /or Cognitive Science

How does Marr's approach contribute to classical debates about visual
perception? In general, he favors a bottom-up analysis of images. Rather
than beginning by hypothesizing about what might be there, the percep-
tual system consists of many specific computational mechanisms targeted
to carry out specific analyses, independent of any "real world" knowledge
or any interaction among the several mechanisms. Indeed, Marr maintains
that there are separate modules for computing such aspects of visual infor-
mation as motion, stereoscopy and color, with each module operating
according to its own principles and having little, if any, interaction with
the others. Marr sought to hold off the importation of knowledge about
possible objects and scenes until as late in the processing of visual informa-
tion as possible. For once one begins to draw on such knowledge (of the
sort presumably used by human beings to interpret scenes), the process of
modeling would become much more complex. As Marr maintained, "Per-
haps we could say that at these higher levels we are beginning to face all
the problems that the linguists have" (1982, p. 313).
But if Marr was sympathetic to the bottom-up approach, so long as
it proved tenable, he was equally an apostle of considerable processing and
considerable symbolic representation during early visual processing. That
is, he had no truck with the notion that perception can occur automatically,
because of a prescribed or predetermined "fit" between the perceiver and
the world. Rather, in the tradition of Helmholtz, Marr noted from the first
305
II I I TowARD AN INTEGRATED CoGNITIVE SciENCE
the extensive amount of recalculation of the initial information that is

necessary for effective perception to take place.
In this respect, Marr was close to his M.I.T. colleague Noam
Chomsky, who had stressed the considerable computation involved in
producing and comprehending language (Chomsky 1984). While
Chomsky spurned the methods of artificial intelligence, and was less
concerned with corroborating evidence from other disciplines, he shared
with Marr a belief in the need for an account of language (or syntax) at
the most abstract level-what is the nature of syntax, and how it is pos-
sible for any mechanism to carry out syntactic operations altogether. It is
not surprising that the two men had considerable respect for one an-
other. Just as Chomsky wished to examine syntax in its pristine form
(uncontaminated by semantics or pragmatics), Marr wanted his analysis
of visual processing insulated as far as possible from the intrusion of
"real world" knowledge. But Marr also sought consistency with what is
known about the operation of the brain. At each level of processing, he
examined the relevant evidence about brain operation: he continually re-
vised algorithms in the hope of making them consistent with the psycho-
physical and neurophysiological evidence-be it at the level of individual
cells or of cortical lobes. Thus, with one eye fixed upon the brain, the
other upon implementation of algorithms on a computer, Marr was an
embodiment of interdisciplinary cognitive science.
In addition to his concerted effort to illuminate visual processing,
particularly during its early stages, Marr also had strong views about how
other research efforts in artificial intelligence should be carried out. He was
critical of those approaches that focus only on mechanisms, that tinker
with a particular problem without understanding the nature of that prob-
lem itself. By the same token, he denigrated those extremely general ap-
proaches which assumed that all kinds of problem solving should be
modulated by the same mechanisms. In information processing, he main-
tained, the structure of each individual problem is central and must be
understood first.
Marr felt it important to focus initially on those tasks humans carry
out well, rather than to examine those they do poorly. Such well-executed
processes were more important for human beings, in general, and more
likely to entail long-standing highly organized mechanisms. In contrast,
the study of tasks (like the solution of cryptographical problems) which
most individuals can carry out only with difficulty would turn up only ad
hoc and artificial-processing strategies: these were unlikely to illuminate
deeply entrenched aspects of human information processing.
Marr's colleague Christopher Longuet-Higgins paid his late colleague
high compliments:
306
If neurophysiology was a theoretical vacuum when he entered it, it is now
seething with lively controversy about the validity of his ideas on the visual
system .... Even if no single one of Marr's detailed hypotheses ultimately survives,
which is unlikely, the questions he raises can no longer be ignored and the method-
ology he proposes seems to be the only one that has any hope of illuminating the
bewildering circuitry of the central nervous system. David Marr's lifework will
have been vindicated when neuroscientists cannot understand how it was ever
possible to doubt the validity of his theoretical maxims. (1982, p. 992)
Their fellow countryman Stuart Sutherland offers even higher praise:

"Vision [Marr's posthumous book] is perhaps the most important book on
the subject to appear since Helmholtz' Physiological Optics" (1982, p. 692).
In a nontrivial sense, Marr was picking up the program glimpsed
during the late 1940s when notions of cognitive science were first jelling.
Workers like Wiener, von Neumann, and McCulloch, who were interested
in the connections between the brain, the computer, and processes of
perception, put forth various proposals about how the perceptual mech-
anisms actually work. When the first conception proved wrong, or impos-
sible to verify, efforts to tie together these sources ceased for a while, and
scholars withdrew into their own specialty: psychologists like Gibson,
focusing on processes of visual perception; artificial intelligence experts
like Frank Rosenblatt directing their energy to computer vision; neuro-
physiologists like Lettvin and Hubel and Wiesel stressing brain neuro-
anatomy. Using the computer but focusing on the theory of perception,
Marr was able to move some steps toward an integration of these different
perspectives. As he articulated his own vision, "The discovery of con-
straints that are valid and sufficiently universal leads to results about vision
that have the same quality of permanence as results in other branches of
science"(Marr 1979, p. 75).
Reactions to Marr
It is too soon to evaluate which portions of Marr's program are likely
to survive, and which will need to be modified or altogether scuttled. Most
commentators have expressed admiration of Marr's research program and
awe at the amount he was able to accomplish with his colleagues in a few
years, but have generally withheld public comment on which particular
aspects of his effort are least robust. Conversations with scholars know-
ledgeable about vision have revealed areas of likely criticism.
Like many other pioneers, Marr expressed his convictions in a strong
and clear-cut manner, even as he did not hesitate to criticize others. In
reaction, not surprisingly, observers stress the ties between Marr's contri-
butions and those of other workers (for example, Hom 1975) as well as the
307
connections among the various levels of analysis that he preferred to

isolate. Apparently not all of the processes described in his writings were
implemented in computer programs; and until they have been, it will
remain uncertain which are workable. There are reservations about how
much visual processing can actually take place without top-down informa-
tion: to the extent that such top-down effects may occur early on, or be
more pervasive in early visual processing, or allow shortcuts to recognition,
Marr's modular approach is endangered. And except at the level of the
primal sketch, little neurological evidence can be related directly to the
Marr model.
Questions have also been raised about the extent to which Marr's
analysis successfully addresses the most central part of perception-the
recognition of real objects in the real world. Most of his account focuses
on the steps before object recognition takes place; the procedures he out-
lined for object recognition may prove applicable chiefly to the perception
of figures of a certain sort-for example, the mammalian body, which
lends itself to decomposition in terms of generalized cylindrical forms.
Like other pioneering researchers, Marr helped to identify the agenda
for future work. A large cadre of scholars are now engaged in interdiscipli-
nary studies of visual perception; and irrespective of their attraction to
Marr's particular claims, they are working on the problems he did much
to bring to the fore. A far more critical perspective on his whole enterprise
can, however, readily be inferred from a group of researchers who rarely
discuss his work: those inspired by the writings of James J. Gibson. The
line of thinking emanating from Gibson serves as an important challenge
to the general cognitive-science approach to perception and to other issues
of human cognition as well.
The Gibsonian View of Perception
In Gibson's view (1950, 1966, 1979; see also Turvey et al. 1981), organ-
isms are so constituted, and live in a world so constituted, that they will
readily gain the information they need to survive and to thrive. In partic-
ular, our sense organs are designed to pick up information from the ex-
ternal world. Thus, when one detects the third dimension, the relevant
spatial information is simply presented in the light, without one's having
to infer distances or to correlate information from eye and hand; there is
no need for the kinds of unconscious inference which scientists from
Helmholtz on have proposed. Initially in life the information will be
308
relatively gross, but with time and experience, it will be increasingly
fine-grained; still, in any. case, the information will be adequate for sur-
vival, veridical, devoid of misleading cues. The information is available
in the world and needs only to be picked up. There is no need to operate
upon it or process it; there is no need to draw on prior knowledge, on
mental models, on interpretive schemata (Gibson 1967). If researchers
understood how people attend to what is there, all the problems of per-
ceptual psychology-indeed, perhaps all the problems of psychology
proper-could be satisfactorily explicated.
On which evidence did Gibson draw in coming up with this radically
simple orientation to perception? In the course of many years of ex-
perimentation, he became extremely impressed with the kinds of informa-
tion available to the visual sense alone. For example, according to classical
empiricist theory, the sense of touch was deemed central to an organism's
ability to perceive depth. But Gibson showed that there is sufficient infor-
mation in the visual sphere to allow the discrimination of depth. The
gradient of texture density is an important clue. Consider a piece of paper
or a linoleum floor with a recurrent checkerboard pattern: one sees the
density of the texture as invariant across the entire array. So long as one
peers straight ahead (or straight down) at the pattern, the density of the
texture does not change. But as the stimulus slants, or as one peers at it
from an angle, the density of texture changes from the near edge to the
far edge, and the viewer receives precise and unambiguous information
about the distance of each part of the stimulus from one's eyes (Hochberg
1978). Hence attention to texture gradient suffices to yield information
about the sizes of objects on a surface or the arrangement of surfaces with
respect to each other: there is no need for the use of other sensory modali-
ties, the calculation of ratios, or other unconscious inferences.
Gibson also stressed the important contribution to perception made
by a person's motions in the world. So long as one is forced to sit passively,
any scene will appear ambiguous. But if you are free to walk about,
changes in the optic array will be precisely tied to the voluntary move-
ments of your body. As you continue to explore, information is routinely
obtained and that information in tum yields more relevant information.
Moreover, the changes in the optic array that result from motions initiated
by the individual make it very easy to figure out what is occurring in the
visual world. Thus, active exploring individuals exploit a perceptual
system that is maximally informative about space and distance.
Based on these and many other observations and considerations, Gib-
son arrived at extreme skepticism about the whole computational ap-
proach. He objected to the notion of mental representations, mental opera-
tions, the processing (as opposed to the direct "pickup") of information, and
309
other cognate concepts. Inferences were completely unnecessary. From his

point of view, cognitive scientists have been subscribing to modes of
discussion and analysis which are both unnecessary and fundamentally
wrongheaded.
In the light of today's psychology, this is certainly a radical perspec-
tive. Indeed, a reader of this book might well ask why, in this cognitively
oriented age, anyone would take it seriously. There are at least three
reasons, I think. One is that Gibson was an extremely clever and incisive
researcher, a keen student of perception, who helped to explicate many
perceptual phenomena and underscored-as Marr would later stress-the
extensive information already present in the environment in which the
organism lives. As Marr put it, Gibson asked the critically important ques-
tion: "How does one obtain constant perceptions in everyday life on the
basis of continually changing sensations? This is exactly the right question,
showing that Gibson correctly regarded the problem of perception as that
of recovering from sensory information 'valid' properties of the external
world" (Marr 1982, p. 24).
A second reason for Gibson's appeal can be found in the set of inter-
locking concepts he developed. He felt that psychologists ought to look to
physics-most especially to optics-as a means of understanding the
structure of the environment. Physics would be linked to biology as organ-
isms make their way in the world, picking up relevant information to
which they are attuned. As a central part of his fine-grained analysis of the
environment, Gibson introduced the notion of alfordances. Affordances are
the potentialities for action inherent in object or scene-the activities that
can take place when an organism of a certain sort encounters an entity of
a certain sort. According to the notion of affordances, individuals throw
things that are grabbable (that is, they afford grabbing), devour things that
are edible (they afford eating), and cuddle things that are lovable (they
afford loving). The concept of affordances permitted an analysis of an
organism's effectiveness within its environment, without the apparent
need to invoke beliefs, attitudes, or mental effort-all of which concepts
made Gibson very nervous. Indeed, for him an object's meaning consists
in the affordances it provides to an organism. Objects are meaningful to
us because they afford things for us to do with or to them or in reaction
to them.
A final reason for Gibson's persuasiveness comes from the simplicity
of his point of view. Like B. F. Skinner, who toiled in a distant area, Gibson
was able to provide an unambiguous theory, which apparently did away
with the need for a lot of internal apparatus and representational contriv-
ances, and substituted a straightforward realistic account, where the world
simply contains whatever information is needed. "I am convinced," Gib-
310
son once declared, "that invariance comes from reality, not the other way
around. Invariance in the ambient optic array over time is not constructed
or deduced; it is there to be discovered" (quoted in Royce and Rozeboom
1972, p. 239).
It is here that cognitive scientists of almost every stripe have locked
horns with Gibson. As Marr phrased it, the detection of invariants is
exactly and precisely an information-processing problem, which must be
approached by the tools of modem psychology. Gibson greatly underrated
the difficulty of such detection. The only way to understand how the
detection works is to treat it as an information-processing problem.
Cognitive-Science Critiques of Gibson

In order to meet the challenge put forth by Gibson and his followers,
the cognitive-science community has rolled out its big guns. Shimon Ull-
man, a colleague of David Marr, mounted an attack "against direct percep-
tion" in The Behavioral and Brain Sciences (1980). Ullman points out that the
direct perception of information touted by Gibson would be justified if it
were not possible to break down perceptual information into more elemen-
tary constituents; but computational and psychophysical studies have
amply shown that the perception of objects can be broken down into
simpler units and into more elementary operations.
In Ullman's view, it thus becomes preferable to posit a continuum of
perceptual mechanisms ranging from direct to indirect. Some aspects of
perception, such as those involving the detection of textures or the map-
ping of a reliable correlation between light and layout features of gradi-
ents, seem to work in a relatively direct way. The field owes much to
Gibson for having identified these perceptual capacities and clarified
their mode of operation. But many other aspects of perception simply do
not lend themselves to direct registration and interpretation. The raw
intensity array which constitutes the first image is unmanageably vast:
the only feasible first step is to replace the intensity array by a represen-
tation of the significant intensity changes in the image, as Marr and his
colleagues have done in positing the primal sketch. Moreover, other as-
pects of perception-for example, the perception of motion-simply do
not occur in a direct way but, rather, involve the devising of structural
descriptions on which perceptual mechanisms can then operate. And by
the time one is dealing with object recognition, one has encountered a
process that is mediated by prior knowledge and by beliefs through and
through and can in no way be accomplished directly by detectors. In
Ullman's perspective, the richness of information in the visual array does
not imply that constructs internal to the perceiver have no place in the
311
theory of perception. On the contrary, it is only through a joint recogni-

tion of the rich information in the environment and of the capacity for
considerable recoding and transforming that we become able to see and
to recognize objects.
Ullman is challenging Gibson on two fronts. First of all, he is saying
that a great deal can be gained by studying perception in terms of its
component parts and phases (rather than by viewing it as an automatic
kind of "resonance" between the perceiver and the world); thus a bottom-
up approach has its utility. Second of all, he is insisting on the importance
of an analysis in terms of an internal symbolic representation: the mechan-
ics of perception simply cannot be modeled or accounted for except in
terms of such a cognitive-scientific approach. In short, the admitted rich-
ness of the perceptual array in no way implies that the absence of con-
straints is internal to the perceiver.
Even more of a frontal attack against Gibson was carried out by Jerry
Fodor and Zenon Pylyshyn in their article "How Direct Is Visual Percep-
tion: Some Reflections on Gibson's 'Ecological Approach'" (1981). De-
fending what they wryly term the "Establishment" view, these two au-
thors single out for special attack the central notion in Gibson's scheme-
that objects have properties or affordances that are resonant with the needs
or goals of organisms and are therefore readily responded to or "picked up"
by the organisms.
The concept of affordances allows Gibson to explain how organisms
know what to do with or how to respond to what they perceive. But
Fodor and Pylyshyn argue that the properties of affordances are uncon-
strained-that anything can be an affordance. But if any property can
count as affordance-and moreover, if any such property can be picked
up-then one is left with an account as empty as Skinner's alleged expla-
nation of language learning. And so, playing Chomsky to Gibson's Skin-
nerian account, Fodor and Pylyshyn conclude that Gibson's explanation
of why organisms handle objects in the way that they do is devoid of
theoretical significance. As Fodor declares elsewhere, "The category
'affordance' seems to me to be a pure cheat; an attempt to have all the
goodness out of intentionality without paying any of the price" (Fodor
1980, p. 107).
As this last quotation suggests, the key concepts missing from a Gib-
sonian account all have to do with meaning-with how an organism inter-
prets, infers, or assumes an "intentional stance" to an object. Only some
such notion will prevent the organism from noting everything and re-
sponding to everything. On the Establishment account, Gibson fails to
appreciate that individuals are able to act appropriately in an environment
because they make inferences about what they see, and because they have
312
beliefs, goals, purposes, and other intentional states directed toward their
percepts. As Fodor and Pylyshyn point out, it is not good enough that we
directly perceive that a rock can be used as a weapon (because it is "grabba-
ble11 and "hurlable11); we need an account of how apprehension of such a
property can occur without inferences. There is no "neutral11 information
available to a perceiver; either the information is in the world but not
available in a relevant form to the perceiver; or the information is inter-
preted by the perceiver and hence can no longer be deemed neutral
(Fodor 1984).
Thus, when Gibson talks about the light conveying information con-
cerning a layout, he is actually talking about semantic relations-the infor-
mation is about something; but he has no way of indicating how one
recognizes such semantic relations. The question how one gets from prop-
erties of light to the properties of the layout has only one conceivable
answer in the Fodor-Pylyshyn view-through inferential mediation. The
pressing question for the cognitivist is to understand the sorts of empirical
consideration relevant to deciding which properties of the light are attended
to, interpreted, and then acted upon.
In Fodor and Pylyshyn's view, Gibson's longtime focus on problems
of visual perception led him to underestimate the difficulty of constructing
a cognitive psychology that dispenses with mental representation. The
prototypical perceptual relations are extensional: they have to do with the
information about features of the environment, such as those in an image.
Here the need for internal representations and inferences may be less
patent. But most prototypical cognitive relations-like believing, expect-
ing, thinking, and so on-are intentional, and the constructs of mental
representation are required to explain intentionality. There is a big differ-
ence between seeing x and seeing x as y, and what is central to cognitive
psychology is the ability to see something as an entity-a rock as a tool
or a weapon or a stool rather than just a blob.
Fodor and Pylyshyn (1981) conclude that one needs either an inde-
pendent account of the meaning of a representation, as is called for by the
Establishment view, or of the specification of a property, as required by
Gibson. The problem of accounting for the meaning of a representation
may be tractable, since the meaning of a representation can perhaps be
reconstructed by reference to its function; but Gibson gives no indication
at all of how to specify a property, no explanation of how a configuration
of light can ever specify such interpreted properties. And so "where the
Establishment line offers, anyhow, a pious hope, the Gibsonian line offers
only a dead end" (Fodor and Pylyshyn 1981, p. 192). Missing the point
about inference, about mental representations, about intentionality are all
thus aspects of missing the same point.
313
II I I TowARD AN INTEGRATED CoGNITIVE SciENCE
An Aggressive Defense
With such a staunch attack from the M.I.T. trio of Ullman, Fodor, and
Pylyshyn, * one might wonder whether the followers of Gibson would
retreat (or retrench) in silence. Far from it. In their commentaries on Ull-
man's target article in The Behavioral and Brain Sciences, and in a response to
Fodor and Pylyshyn even longer than the original article, Michael Turvey
and Robert Shaw, two Gibsonian psychologists at the University of Con-
necticut, indicate that the cognitivists have missed the point of their own
position. To start with, Turvey and Shaw remind us of the extent to which
Gibson's work has illuminated fundamental questions of perception and
has helped to explain the perception and the behavior of a wide range of
organisms moving about in a wide range of environments. The proof is in
the pudding, they say, and they reject out-of-hand the notion that the
Gibson approach is as vacuous as Skinner's: they then wrap the mantle of
Chomsky around themselves by claiming, "Just as Chomsky used the
regularity and ease of natural language acquisition as a fact to justify
treating language as a special subject matter, so Gibson and his followers
have argued for the importance of doing justice to natural, effective
perception" (Turvey et al. 1981, p. 239).
According to Turvey, Shaw, and their assoc_iates (abbreviated here-
after as Turvey and Shaw), Gibson's ecological approach seeks to account
for an organism's apprehension of its environment and the manner in
which it controls its actions with respect to that environment. The ap-
proach focuses squarely on organism-environmental relations, never on
what might be in the organism's head. Indeed, many issues can simply be
bypassed: for Gibson, "awareness" is always "awareness of some prop-
erty," and so there is no need to posit inferences or symbolic calculations.
Secondary become the issues of what counts as perception, of whether
perception should be construed as judgment, of whether perception is
direct or indirect, and of how inference figures in the scheme of things.
Once again turning Fodor and Pylyshyn's argument against them, the
Turvey team concludes that the Establishment's views are insufficiently
constrained and have placed a burden on inference which it simply cannot
bear. Instead, Turvey, Shaw, and colleagues argue for a conception of
natural law that posits meaningful relations between organisms and envi-
ronment. By "natural law," they are referring to scientific principles-in
this case, laws that explain why organisms perceive and behave in the
ways they do by virtue of their fit with their environment. The use of the
term perception should then be restricted to relations captured by such laws.
*Pylyshyn, currently at the University of Western Ontario, is a frequent visitor and
sometime collaborator at M.I.T.
314
Ecological theorists like Gibson seek to extend the application of
natural law as far as possible, because that strategy promises tight con-
straints and an explanation of lawful evolution of inference in principled
ways. In contrast, Establishmentarians like Fodor and Pylyshyn want to
extend cognition and intelligence as far as possible and thereby to limit the
role of natural law. Regularities, then, must be accounted for by mental
rules and representation-which are themselves constrained by very little
except unsystematic intuitions (Turvey et al. 1981, p. 245). The Establish-
ment talks a great deal about how to make the right inferences and very
little about how to locate the right premises. The ecological school provides
a way to get the right premises by laying out the laws of perception; it
derives these laws of nature by careful observation of, and experimenta-
tion on, organisms as they pick up information from and operate effec-
tively within their natural environments. The Establishment has to gamble
that it can guess how an organism is going to know what to do in a given
situation; the ecological school has the much more straightforward task of
simply finding out what sorts of things this kind of organism actually
does in this kind of a situation.
The role of natural law centers on affordances and effectivities. Gib-
son's "ecological science" studies the relations between affordances-things
that are grabbable, climbable, and the like-and effeclivities-things that can
do the grabbing, climbing, and so on. The examples used by the Gibson
school account for how organisms carry out these and countless other
activities in their natural environments in order to survive and thrive.
According to this scientific credo, psychologists are called upon to describe
the lawful regularities of the environment if they wish to produce a scien-
tific explanation of the origin, function, and causation of behavior (Turvey
et al. 1981, p. 274). This tack includes laws about everyday objects and
events, which have the affordances that govern behavior of importance to
the organisms of the world. And so, the "ecological psychologists" fill their
articles with analysis of places where wasps can lay their eggs, elements
that sharks can eat, stems on which marsh periwinkles can climb, and the
like-no optical illusions or ambiguous sentences for them.
Turvey and Shaw deplore the rampant tendency to ascribe to an
organism neural detectors or structural descriptions that essentially repre-
sent the very property that is sensed. This practice permeates the Estab-
lishment: for Z to see, detect, register a property x of X, Z must "have"
property x in some sense, neurophysiologically or conceptually. In contrast
(as Turvey and Shaw have put it), the ecological school eschews rules (or
computation) in favor of natural laws, representations in favor of occurrent
properties, and concepts in favor of affordances. Instead of trying to stuff
properties into the heads of organisms, the ecological school keeps these
315
III I TowARD AN INTEGRATED CoGNITIVE SciENCE
properties where they belong-in the natural world. All told, the ecologists
make central to their science the discovery of natural laws governing the
organism's relation to its environment, rather than cognitive laws, which
carry out operations on mental representations.
Contrasting Perspectives
As the ecologists describe it, we are left with two sharply contrasting
images. The Establishment treatment of the intentionality problem con-
jures up an image of an organism (such as the hermit crab), on the occasion
of becoming hungry, moving about with a concept of food in mind and
looking for something that will match the concept; or an organism (such
as the marsh periwinkle), on the occasion of impending danger from the
approaching tide, moving about with a concept of a thing that can be
climbed up in mind and looking for something in the environment that will
match up with that concept. The ecological approach conjures a very
different image-of an organism, on a given occasion, moving in the con-
text of one set of nested laws rather than another. Confronted with the fact
that a gannet can dive accurately into the water and scoop up a fish when
hungry, the ecological school asks about the various kinds of laws that can
control the gannet's behavior and how the values for the gannet's dive are
arrived at in accordance with some law when it dives for food. The latter
image expresses belief in a natural basis to intentionality, whereas the
Establishment image does not (Turvey et al. 1981, p. 299).
In the view of Turvey and Shaw, there is at stake a larger philosophy
of science issue. Are the uniformities observed in nature reflections of an
underlying concrete framework of laws, or are they only the insidious
invention of the human mind? Turvey and Shaw criticize Kant's arrogance
in saying, "The understanding does not draw its laws from nature but
prescribes them to nature" (1981, p. 299), and prefer to draw constraints
from biology and physics rather than from the more elusive realm of the
human mind. These researchers prefer to believe that there is a natural
order, which it is the scientist's role to discover-just as it is the organism's
role to discover the natural properties in its environment that will give it
all that it ever needs to know.
The Gibsonians attack the Establishment in terms of its conception of
scientific truth. While both Gibson and his critics are involved in a scien-
tific effort designed to figure out how perception and cognition occur, the
Gibsonians are likely to find the crucial information in the environment
316
and the organism's relationship to it; the Establishment cognitivists pay
attention instead to the presuppositions and biases built into the organism
and to the way information in the world becomes transformed or recon-
structed upon its apprehension by the organisms.
It is possible to take the position that these are simply two views of
the same basic situation: one paying far more attention to the environment;
the other paying far more attention to the organism; but both attempting
to explain the same set of phenomena in parallel ways. One can also divide
the territory to be explained, leaving simple perception to the ecologists,
complex inferencing to the Establishment. Both parties occasionally take
such positions, when they are in a mediating mood; but in general, the gulf
seems much deeper. The very terms that are central to the Establish-
ment-in/en/ions, inferences, schemas-are rejected by the Gibsonians as un-
necessary: even as the terminology favored by the latter--alfordances, elfecHoi-
fies, in/ormation pickup-is deemed by the Establishment to be vacuous or
insufficient. The fundamental scientific themata are also different: Gibson
reflects a belief in the real world as it is, with all the information there, and
the organism simply attuned to it; the Establishment reflects a belief in the
constructive powers of the mind, with the external world simply a trigger
for activities and operations that are largely built into the organism. De-
pending upon whose view holds sway, cognitive studies will look quite
different.
From one vantage point, the dispute between the Establishment and
the ecological school can be depressing. Here we are, two thousand years
after the first discussions about perception, several hundred years after the
philosophical debates between the empiricists and the rationalists first
raged, and leading scientists are still disagreeing about fundamentals.
Though the current debate cannot be mapped directly onto other debates
-nominalist versus realist, empiricist versus rationalist, unconscious in-
ference versus sensory registration versus "pickup" of relevant informa-
tion-the themes are familiar enough, and the arguments frequent enough,
as to make one question whether there has been progress.
Yet once one transcends these statements of position-in very spare
form-one finds far more agreement. Both schools certainly believe that
our knowledge of the mechanisms of perception has been enhanced over
the past decades. We know much more now about how perception of the
world comes about than we did fifty years ago, and a good deal of credit
for this knowledge accrues to Gibson himself. Disagreements about a
research program lie in whether the study of organisms and environments
can suffice or whether an additional layer of analysis is necessary. Marr and
his colleagues have made a strong case for the proposition that perception
can only be understood if the nature of the problem involved in vision is
317
I I I I TOWARD AN INTEGRATED COGNITIVE SCIENCE
analyzed explicitly-an approach requiring logical and mathematical con-

siderations-and if attempts are made to try to realize these considerations
in terms of a well-specified model.
It seems to me that this question can be-and likely will be-resolved
by future research: by ascertaining whether the Gibsonian approach is
capable of handling more complex aspects of scene recognition and object
recognition, or whether it requires the kinds of inferential and intentional
mechanisms postulated by the Establishment. Or, to put it the other way
around, the conclusion will depend equally on whether the Establish-
ment's promissory approach actually can help us to understand the gannet
in flight or the infant moving about its playpen.
Possible Reconciliations
Neisser s Ecological Approach

Recently two trends, occurring in contrasting comers of the cognitive
sciences, hold promise for the reconciliation between the two perspectives
outlined here. One trend embodied in the recent work of Ulric Neisser
(1976, 1984) is clearly in a top-down direction. Once the prototypical
cognitive psychologist, Neisser has become increasingly impatient with
those artificial laboratory studies and information-processing accounts
that have become the mainstay of his discipline. Following some years as
Gibson's colleague, Neisser has become a moderate convert to the ecologi-
cal stance. The form of this conversion consists less in a doctrinaire en-
dorsement of Gibson's particular concepts than in a commitment to study
the kinds of behavior that interested him under the "real world" condi-
tions he favored. Thus, as noted in chapter 5, Neisser calls for studies of
perception as it unfolds when an organism is making its way around the
world; studies of recognition and classification of complex real world ob-
jects as they are encountered in the environment, rather than of contrived
objects encountered in lab settings; and studies of memory of one's early
history or of complex experiences in the natural world. Neisser hopes that,
from a wedding of the concerns of cognitive science with the naturalistic
approach laid out by the Gibson school, one may end up with a science
fully equal to the species.
Parallel Processing in Perception

The other recent trend, which may help to moderate the Gibson-
Establishment dispute, is distinctly in the bottom-up tradition. I have in
318
mind a new wave of artificial-intelligence attempts to study the process of
visual perception. Inspired by Marr's example, but diverging from it in
significant respects, a sizable number of scientists are now investigating
visual-processing systems modeled closely on the primate nervous system
(Ballard, Hinton, and Sejnowski 1983; Brown 1984; Feldman 1981; Gross-
berg 1980; Hinton and Anderson 1981; Hinton, Sejnowski, and Ackley
1984; Hofstadter 1983; Rumelhart and McClelland 1982; Waldrop 1984).
These approaches-variously called "parallel visual computation," "nee-
associationism," "nee-connectionism," or "massive parallel processing
systems" (M.P.P.S.)-begin with a critique of standard computer simula-
tion of vision, which uses serial"von Neumann-style" symbolic process-
ing. In their view, it makes far more sense to simulate vision by using
(physical or virtual) machines involving many independent processors
(perhaps as many as a million) carrying out many processes at one moment
in time. Such parallel systems take the Marr scheme much further than he
explicitly went. To the notion of different modules carrying out their own
separate analyses, one now adds the notion of many units operating and
exchanging information in ways analogous to many brain cells or columns
firing simultaneously. Computation is performed by excitatory and inhibi-
tory interactions among a network of relatively simple neuronlike units,
which compete and cooperate so that certain units become active and
others are suppressed. Eventually, thanks to statistical properties of the
ensemble, the network settles into a state that reflects its particular "task"
-for example, perceiving a given image.
In these M.P.P.S.'s, memory and perception occur in a distributed
fashion: that is, instead of there being a single central control, or the
complex passing of information between modules, many units operate
simultaneously and achieve their effects statistically. The multiple connec-
tions allow much of the knowledge of the entire system to be applied in
any instance of recognition or problem solving. There are other advantages
to a distributed representation. No information inheres in a specific locus;
thus, even though many units (or cells) may be destroyed, the relevant
memory or concept continues to exist. Also, because of the widespread
distribution of information, it is possible to arrive at a decision even if a
match turns out to be noisy, or incomplete, or approximate. These proper-
ties seem closer to the kinds of search and decision organisms must carry
out in a complex and often chaotic natural world.
An important feature of the M.P.P.S. approach is that it dispenses
with some of the staples of artificial intelligence as the latter has been
customarily conceived (and as it most annoyed Gibson). Central to the
classical Newell-Simon view is the positing of symbolic structures, upon
which operations are performed in a specified order, as the result of a
decision procedure. In this new dispensation, it may be possible to dispose
319
with any notion of symbolic processing and, instead, to model perception

in a more direct fashion-just neurons connected to other neurons and
dedicated to specific functions. Instead of the need for a separate knowl-
edge store, knowledge (and even "intelligence") simply inheres in the
strength and appropriateness of the connections between simple neuron-
like processes. And, indeed, the machine concepts involved in this new
approach bear provocative analogies to the primate brain.
In some ways, the parallel-processing approach seems a throwback to
the first days (or even the gestation period) of cognitive science. The kinds
of neural network being simulated resemble those that first drew McCul-
loch and Pitts's attention to the relations between the nervous system and
the logical calculus. The desire to model the brain directly is reminiscent
of early attempts to build computers that realized Hebb-type cell assem-
blies and phase sequences. The statistical"pandemoniumlike" aspects of
these models and the clear absence of controlling "executive processes"
contrast with the logical step-by-step difference-reducing programs that
came to be perfected by Newell and Simon in their General Problem
Solver. Yet, because of the greater rigor with which the networks are
specified, and enhanced understanding of the problems involved in image
processing, these approaches may be viewed as a tentative synthesis be-
tween the classical and the Marr approaches, rather than simply as a
regressive force. Indeed, the facts that several of the algorithms have
successfully simulated aspects of visual perception, and that the operation
of these systems seems consistent with what is known about the operation
of the primate visual system, give hope that such parallel systems may
eventually serve as reasonably complete simulations of early visual per-
ception.
In spite of the recent excitement about massive parallel processing
systems, they leave questions unanswered. While these systems can ac-
complish much more in a short period of time, speed alone does not ensure
understanding: the fundamental questions about the nature of perception
(and of other systems) must continue to be raised and addressed. Basic
computational and definitional issues do not vanish just because of speed
or brute force. There remain unresolved questions about how the numer-
ous independent processors manage to work together and whether they
can simulate behavior that is truly sequential-for example, eye move-
ments. As in Gibson's work, questions also arise concerning the extent to
which mechanisms of this sort can be adapted to more complex forms of
problem solving, of the sort that has classically been simulated by symbol-
manipulating von Neumann machines.
In fact, some authorities have raised the intriguing notion that the
brain (and hence the computer) might be most effectively thought of as
320
entailing two different systems: one, massively parallel, engaged in such
probablistic endeavors as object recognition: the other, sequential, dedi-
cated to symbolic manipulation, rendering dichotomous judgments, and
engaged in such deterministic activities as logical problem solving (Kosslyn
1984; Fahlman, Hinton, and Sejnowski 1983).
How does the new approach of M.P.P.S. speak to the issues being
debated by the Gibsonians and the Establishmentarians? Paradoxically,
the parallelists are using the mechanisms most clearly associated with the
Establishment view-powerful electronic computers-in order to put forth
a view of perception closer to that embraced by Gibson. He did not live
long enough to pass judgment on this new approach to the simulation of
vision. But several features of the new approach-its fidelity to the me-
chanics of the brain, its spuming of complex symbol manipulation or
intricate decision procedures about what step to carry out next, its impor-
tation of vast amounts of real world knowledge, and its suggestions about
how Gestalt phenomena might emerge from the competition and coopera-
tion of various neural networks-have a Gibsonian ring to them. More-
over, to the extent that such efforts actually result in an effective simula-
tion of vision, in a way that seems concordant with the human nervous
system, the need for theoretical discussions about the "right way" to think
of perception may become increasingly academic. (But see Fodor 1984 for
an alternative view.)
In any case, if either of these rival schools-or one like them-turns
out to be successful, then the problems of visual perception will have been
largely resolved. The Marr-Ullman approach will represent more of a
triumph for cognitive science: it is a self-constructed dialogue among sev-
eral participant disciplines, no one of which can hold the answer to the
puzzles alone; and it draws explicitly on cognitive science concepts such
as representation and symbolic operations. If the Gibsonian approach were
to triumph, it would call into question the need for much cognitive-science
baggage; it would constitute more of an explanation in a classical psy-
chobiological or psychophysical framework. The M.P.P.S. approach falls
somewhere in between: the technology of cognitive science in the service
of a view of perception which is closer to neurology than to psychology.
In my own view, Gibson's work is a logical place in which to begin
the study of perception: his ecological perspective provides vital informa-
tion concerning the phenomena of perception and the kinds of information
in the environment to which any perceptual apparatus must be sensitive.
In Marr's terms, Gibson helps us to understand the nature of the computa-
tional task involved in perception. But Gibson's work proves of limited
help in understanding the actual steps involved in perceiving specific ob-
jects. In contrast, Marr has put forth a plausible account of how an orga-
321
nism may actually proceed from intensities of light to the parsing of the
objects in a scene. He has provided a mechanistic account of perception
which seems internally coherent and consistent as well with converging
evidence drawn from several cognitive disciplines. And even if his particu-
lar account turns out to be wrong or incomplete, or if the approach of the
parallel processors carries the day, he has defined the likely parameters for
future debates about early visual perception. So, in the terms I have used
in this chapter: even if Marris wrong about the extent to which bottom-up
analysis can occur before top-down factors have to be invoked, he has
made a persuasive case for the importance of the symbolic or representa-
tional level of information processing. As I see it, the burden of proof is
now on the Gibsonians.
Even if Marr (or the parallel-processing school) is right, and we can
account for the perception of objects with relatively little recourse to
knowledge about the external world, still it is patent that the recognition
of objects, or even groups of objects, marks just the beginning of the
cognitive enterprise. Any student of cognition must be concerned with
what happens to these initial perceptions when they are invoked in plan-
ning, problem solving, or simple recollection: one must concern oneself
with the way in which organisms determine the identities of objects, how
to classify them, and how to reason about them. These even more chal-
lenging puzzles will occupy the following chapters.
322
11
Mental Imagery:
A Figment of
the Imagination?
Now we have already discussed imagination in the treatise
On the Soul and we concluded there that thought is impossi-
ble without an image.
-ARISTOTLE
Introduction: Images through the Ages
Assuming a behaviorist pose vis-a-vis other organisms makes a certain

sense. After all, since one cannot have access to the internal experiences
and sensations of an animal or another person, it may be wise to suspend
judgment about these experiences and to concentrate instead on activities
that are manifest or at least readily measurable. But when contemplating
one's own mental processes, exclusive perseveration on overt behavior
seems unjustifiable. Apsychologist may be disinclined to attribute images,
hallucinations, or dreams to other individuals, but it seems self-delusion
-a doubt even Descartes would have spumed-to deny these in one's
own phenomenal experience. And then, once one has accepted one's own
images, a new puzzle arises: Can one continue to withhold them from
others?
323
It is not surprising, therefore, that most students of mental processes,

from the time of the Greeks on, have singled out for study the capacity
to conjure up in one's "mind's eye" various objects, scenes, and experi-
ences--entities that are not present in one's surroundings (and may never
have existed at all). At first philosophers, then armchair psychologists, and
finally the first generation of experimental psychologists were all keen to
explore mental imagery.
Why this interest in imagery? Seeing or touching the world of objects
can be fascinating, once one begins to think about it, but a person could
certainly pass through life without giving such experiences a further
thought: they are as evident as the prose unknowingly spoken by Moliere's
bourgeois gentleman. But when it comes to experiences that, however
vivid to oneself, are not accessible to other individuals, intriguing ques-
tions arise. It is easy to understand why one should be equipped to react
to things in one's environment-how else would one survive?-but what
end is served by experiences that are completely private? Is this process
adaptive or can it be damaging? Why does one dream? Can one imagine
anything at all and, if not, what are the limits? Does imagery occur in the
same way as normal perception, merely activating perceptual structures by
internal means instead of by a stimulus--or does it marshal different
mental processes? To what extent can one exert control over one's imagery
-and is it possible to influence the imagery of others?
The first psychologists made the study of internal imagery central in
their discipline. Followers of Wundt probed their own internal imagery
and painstakingly analyzed self-reports by their trained subjects. And yet
hardly a quarter-century had elapsed before grave doubts were raised
about the status of mental imagery as a partner fit to participate in polite
psychological publications. Thanks to the skeptical accounts emerging
from Wiirzburg, the more ephemeral and less reliable aspects of imagery
were underlined. Not everyone had images, it turned out, and those who
did introspected about them in different ways. There was no reliable way
to define imagery in an experimental situation, no agreement about what
should count as an imagistic or imaginary experience. Certainly a new
science ought not to embrace such a vague and fuzzy concept as a principal
mental construct, let alone as an explanation of the way in which people
think. For such reasons, the ghostly image was exorcised for half a century
from respectable academic psychology.
In the early 1970s, however, with behaviorism on the wane, psycholo-
gists began to report findings that were difficult to think or write about
except in terms of imagery (Paivio 1971). Perhaps the most dramatic stud-
ies were carried out by Roger Shepard and his colleagues at Stanford
University. In an oft-cited study, which I reviewed in chapter 5, he and
324
Mental Imagery: A Figment of the Imagination?
Jacqueline Metzler (1971) exposed subjects to geometric figures and asked
them to indicate as rapidly as possible whether the two figures were
actually representations of the same object as seen from different vantage
points.
But, strikingly, the difficulty of the task (as measured by the time it
took to give a response) proved readily predictable in terms of the number
of degrees by which the second figure had been rotated. Thus, a figure that
had been rotated eighty degrees took longer to specify as identical to the
target than one rotated fifty degrees, but a model rotated one hundred
degrees proved even more difficult to specify as identical to the target than
had the former two. It was tempting to conclude that the subjects were
mentally rotating these figures, and that the greater the distance of the
rotations, the longer the road to a correct response. Moreover, there was
no need just to conjecture, because the subjects themselves verified this
account. It seemed reasonable--indeed, almost inescapable--to conclude
that human beings generate mental images of these forms and rotate them
through some as yet undefined mental space.
According to Stephen Kosslyn {1980), possibly the major contempo-
rary student of imagery, Shepard's findings caused a sensation in the
cognitive community. The data were striking and systematic, apparently
reflecting a basic capacity of the cognitive system. Moreover, these elusive
internal images turned out to yield a psychophysical law that was simple
yet robust: time to judge identity (or non-identity) was a monotonic func-
tion of the physical distance between the forms. One of the least tangible
constructs in psychology had yielded a scientific law of striking precision.
There were other reasons for the excitement. In light of Shepard's
results, it made sense to think of an individual as having pictures in his
head. He never employed this expression, which cannot be construed in
a sensible manner anyway; but he made respectable the idea of an analogue
mode of mental representation, a mode capturing certain of the relations
of proximity that can also be perceived in the physical world. Shepard's
results called into question current efforts to explain all of thought in terms
of one kind of computational mechanism-that of the serial, digital com-
puter which processes one kind of information. The typical approach of
this era had held that information is represented in the brain in lists or
networks of propositional information. But this approach-that what we
know consists of lists or of propositions-seems utterly inadequate to
account for the mental rotation findings and phenomena. Instead, it made
more sense to think of the brain as passing through an ordered series of
states, which mimic ordinary processes when physical stimuli are being
observed in the world. Imagery should be thought of in its own terms,
rather than as a cryptic version of verbal mediation or symbolic manipula-
325
tion. Perhaps there were two separate and equally valid forms of mental
representation. And perhaps a study of the less familiar imagistic mode
might help to clarify some of the later stages of visual perception-and do
so in a way different from either a neurological or an ecological perspective.
Stephen Kosslyn' s Model
Taking off from these and other equally beguiling findings, Kosslyn and
his colleagues have undertaken three major and coordinated lines of work
(Kosslyn et al. 1979). First of all, they have added to the file of empirical
evidence that favors the existence and flexibility of mental imagery. In a
set of extremely ingenious experiments, the Kosslyn team has delineated
major properties of the imagery system.
To convey the Kosslyn approach, let me introduce a representative
study (Kosslyn, Ball, and Reiser 1978). Subjects are shown a map contain-
ing seven fictional locations-a rock, a tree, a beach, a patch of grass, a well,
a hut, and a lake. After having had a chance to familiarize themselves with
the map, one is asked to imagine the map in one's mind and then to answer
various questions about it. For example, one is asked to focus on one
location in the map and then to look for a second one. Assuming that the
second location is on the map, a subject is directed to picture a little black
speck moving as rapidly as possible from the first to the second point and
then to push a button. When one cannot find the named second location,
one is asked to push a second button.
Kosslyn found that the time to scan from location A to location B was
a linear function of the distance between the two sites on the map. Appar-
ently the subjects were scanning a mental image, just as one would scan
the physically present map. As a check that the subjects were not simply
using a propositional representation or a mental list of the locations, sub-
jects were seen in a second condition. Now one was simply asked to
indicate whether a named location was on the map. It turned out that,
under this condition, there was no hint of a distance effect. Evidently
subjects just consulted a list in this instance. Thus, it seems most unlikely
that subjects were employing some propositional representation in the first
study: for, if they had been, the results on both experiments should have
been identical (Kosslyn 1983, p. 46).
What of other results demonstrated by the Kosslyn group? It takes
more time to survey an image of a large object than an image of a small
one; more time to survey an image that travels in a third dimension than
326
Menta/Imagery: A Figment of the lmaginaHon?
one that only cuts across a flat plane; more time to see small details on an
image than large ones; and it is more difficult to imagine objects known
to be hidden or concealed by a barrier than ones known to be in a visible
position. Ever sensitive to the possibility that subjects may be reasoning
(rather than imagining) their ways to the correct answer, Kosslyn and his
colleagues have paid particular attention to experimental conditions where
subjects could not have known the correct answer. They have discovered
that subjects perceive imagistic effects that even scientists could not pre-
dict from a verbal description of the task, but that individuals placed in
the actual concrete situation will reliably experience. On the other hand,
there are some effects-for example, certain afterimages-that individuals
will perceive when confronted with actual displays but that turn out not
to be experienced when one is asked to imagine the customary eliciting
circumstances. These latter lines of research support the independent va-
lidity of imagery. If individuals always behaved as their knowledge dic-
tated-using images just in those cases where they thought that they should-
then they would not display such paradoxical effects.
Having demonstrated a massive number of results putatively due to
some form of mental imagery, Kosslyn and his colleagues have gone on to
devise a comprehensive theory of this capacity. While not boldly claiming
that human beings have pictures in their heads, these researchers defend
the notion of a "quasi-pictorial" form of mental representation called
"imagery." In their view, this form of mental representation is as important
for an understanding of cognition as is the more usually invoked proposi-
tional form.
Here the plot thickens. Many psychologists would be content to allow
imagery back into the psychologist's lexicon and even permit it to be used
as a "local explanatory construct" for certain reliable findings. But when
talk shifts to imagery as a basic property of human cognition-as a primary
way in which information can be symbolized or represented-then psy-
chologists and other cognitivists grow much more cautious. After all, if as
elusive a concept as an image is allowed to serve as a psychological expla-
nation, then on what basis can other "pretenders" be excluded? For their
part, philosophers also begin to get nervous: What does it mean "to have
an image" or to read information off an image? What can an image in the
head be anyway? (Block 198la).
According to the Kosslyn account, images have two major compo-
nents: the surface representation is the quasi-pictorial entity in the active
memory that is accompanied by the subjective (Kosslyn's term) experience
of having an image. The images are likened to displays produced on a
cathode-ray tube by a computer program operating on stored data. In other
words, the images are temporary spatial displays in active memory that
327
are generated from more abstract representations housed in long-term

memory.
These initial abstract representations consist of propositions and other
kinds of non-imagistic information, such as that embodied in concepts.
Thus, in the generation of imagery, there is an interplay between descrip-
tive (languagelike) and depictive (picturelike) memories. But the quasi-
pictorial image is not merely an epiphenomenal concomitant of more ab-
stract nonpictorial processing. Rather, by drawing on long-term memory,
one can generate images, chunk them in various ways, subject them to
various transformations, and classify them in terms of semantic categories
(for example, a particular imagistic configuration can be parsed as a per-
son's nose or as the state of New Hampshire). Kosslyn speaks of a mind's-
eye-like device which is necessary to interpret images or parts thereof:
indeed, in his view, any representational system must include some sort
of interpretive device (Kosslyn 1978). And he stresses that information is
stored in images that are not languagelike but "bear a non-arbitrary corre-
spondence to the thing being represented" (1981, p. 46). In sum, Kosslyn's
model includes a cathode-ray tubelike display medium, techniques for
forming an image on the display, and techniques for interpreting and
transforming information in such a display.
Computer SimulaHon
Nowadays, in psychology, when one has a detailed description of a
process, it is desirable (as well as fashionable) to attempt to develop a
computer model of it-as Kosslyn and his collaborators have set out to do.
They posit two kinds of data structures. The first consists of a surface
matrix which represents the image itself. This image-data structure is repre-
sented by a configuration of points on a matrix. The matrix corresponds
to a visual buffer, a spatial medium used to support the representations that
make possible seeing during perception as well as during imaging. A quasi-
pictorial image is achieved as a person selectively fills in certain cells of this
matrix. The image depicts information about spatial extent, brightness,
and contrast; it has limited resolution, causing contours to be obscured if
an object is too small. The matrix corresponds to short-term visual mem-
ory; and, therefore, representations within it are transient and can be
maintained only through effort.
The second set of data structures consists of the long-term memory
files, which represent the information used in generating images. One kind
of representation stores visual information about the literal appearance of
an object, such as how something looks; included is information on the
object as viewed in a number of different ways. The second type of long-
328
term representation is a set of facts about the imaged objects. These facts
are represented discursively, in a propositional format: included is such
information as the names of the most highly associated superordinate
categories and a classification of the object's size. These sets of data struc-
tures are then used in three imaging processes: routines for generating
surface images, for classifying these images or parts thereof, and for trans-
forming images.
In a highly schematic example, Kosslyn and his colleagues illustrate
how these data sources and processes get mobilized in a program. For
example, in generating a detailed image of a chair, the IMAGE procedure first
constructs a skeletal image of the chair and then searches for factual
information stored in long-term memory for the names of parts that go in
the chair. It may find, for example, that chairs have cushions, and that
these are the foundation part of the chair. PUT and FIND procedures then
locate the relevant part of the image (the seat) by means of a set of
procedures describing the seat. Once the seat has been found, a FIND opera-
tion passes back Cartesian coordinate information of where this part fits
the image. Once the location and size are computed, an appropriate PICTURE
routine is called, and the part is integrated into an overall image (Kosslyn
et al. 1979, p. 542).
A chief reason for actually programming the computer, as opposed to
merely producing flow charts, is that it helps one to discover the conse-
quences of some claim and to study interactions among components (Koss-
lyn 1980). The fact that the simulation works also solves the homunculus
problem: the program performs the integration, and there is no need for
a "little man" to read the images. In some cases, Kosslyn's simulation has
actually brought to light problems that were not anticipated, and has
accordingly inspired a revision of the model. Thus, for example, in the
original simulation there was no provision for scanning to the inactivated
region of the surface matrix. Only when the program was asked to find an
object in a certain image, and received an error message, did Kosslyn make
needed changes in the program. In addition to calling attention to deficits
in the model, the computer simulation has also suggested various new
studies; these have led to further changes in the model which could be
incorporated into further revision of the simulation as well.
In many ways, the Kosslyn program can be considered a prototypical
cognitive-scientific enterprise. First of all, it addresses a long-standing set
of philosophical issues via a systematic program of experimental research.
It deals unabashedly with the level of mental representation, while avoid-
ing many of the weaknesses of earlier attempts to posit and utilize con-
structs of mental imagery. Rooted in the first instance in psychology, it
relies heavily on a model of computer simulation. In addition, as we shall
329
see, there are also significant ties to work in neuroscience and in philoso-
phy, and these are being vigorously explored at present. Many aspects of
image generation-including its similarities and differences from ordinary
perception or ordinary memory-have been clarified. It is not that the
Kosslyn group has resolved all outstanding issues about imagery: in fact,
much of the work and many of the claims remain controversial. Yet,
building on the work of Shepard, Paivio, and other researchers, Kosslyn
has made the study of imagery a respectable topic within cognitive science
and illuminated crucial facets of this mode of mental representation.
Imagery is now central in any cognitive map of the discipline.
The Debate about the Kosslyn-Shepard Perspective

Probably no research in recent cognitive studies has generated so
much controversy as work on imagery. And because Kosslyn has been
eager to talk of mental imagery as a form of representation, and has
devoted much energy to the development, revision, and popularization of
his model, discussion has focused on his own claims. Indeed, not only was
there a major "target" article in The Behavioral and Brain Sciences (1979) on
Kosslyn's work, complete with twenty-five responses by leading cognitive
scientists, but several major philosophical works have reviewed imagery
in considerable detail (Block 1981a; Dennett 1978; Fodor 1975).
Like any contribution that attracts widespread attention, Kosslyn's
work has been criticized at almost every level, from the naivete of the
theorizing, to the mystique surrounding mental imagery, to the possibility
that the experimental sample includes only those subjects who are sug-
gestible, influenced by the instructions, or liable to confabulate their testi-
mony about imagery. (See the response to Kosslyn et al. 1979 in The
Behavioral and Brain Sciences, as well as Kosslyn 1980 and 1983 and Pylyshyn
1984.) Kosslyn, in turn, has written extensively on each of these lines of
criticism and has, in the opinion of many observers, given satisfactory
responses.
Some of Kosslyn' s responses have taken the form of studies that speak
to specific criticisms. For example, he was criticized because his model did
not handle three-dimensional aspects of images. His student Steven Pinker
proceeded to show that, having surveyed a three-dimensional scene, sub-
jects imagined in three dimensions: that is, the time taken by subjects to
scan among objects decreased in proportion to the actual three-dimen-
sional distances between the objects and not in proportion to the two-
dimensional distances captured in a photographic representation of the
scene (Kosslyn 1983, p. 154). Kosslyn was also challenged because his
subjects may have had previous propositional knowledge about some of
330
the effects they reported: in this case, imagistic explanations need not be
invoked. Accordingly, Ronald Finke, another associate of Kosslyn, studied
a phenomenon-the perception of complementary colors as afterimages-
about which nonpsychologists are ignorant (Finke and Schmidt 1978).
Despite their ignorance, subjects asked to imagine the phenomenon re-
ported precisely the same afterimages as do na"ive individuals confronted
with a "real life" presentation. Finally, Kosslyn's overall position has been
bolstered by Martha Farah's finding (1984) that at least one of the compo-
nents of his model-the image-generation component-can be destroyed
in isolation by damage to the brain. Moreover, and in surprising contradic-
tion to what the literature on brain laterality suggests, the image-genera-
tion component seems to be found in the posterior portions of the left (and
not the right) hemisphere of the human brain.
There have been other criticisms as well. Some scholars have ques-
tioned the meaning of key terms in Kosslyn's model and have voiced the
charge of theoretical looseness. For example, the term quasi-pictorial, as
applied to an image, seems a clear case of wanting to exploit the connota-
tion of picture without paying the price for it. There have also been general
criticisms about the ecological validity of Kosslyn's research program. Ulric
Neisser expresses this point of view:
Why does the theory suggested here strike the reader as clever rather than
insightful, as cute model making rather than serious psychology. I think it is
because the thinking of Kosslyn and his collaborators is completely detached from
everything we know about human action or perception.... It attempts to "account
for" a sharply restricted body of experimental results (usually reaction latencies)
by relating it to an equally restricted class of models (usually computer programs
of something similar). (Quoted in Kosslyn et al. 1979, p. 560)
Perhaps because of the fertile experimentation emanating from Koss-

lyn's laboratory, the most telling lines of criticisms do not focus on the
phenomena or the findings, which seem reasonably robust, but are rather
directed toward his claim that there exists a separate form of representa-
tion called imagery, that it exhibits its own properties and operates inde-
pendently of the canonical form of representation in propositions. Some
commentators, like John Anderson and Philip Johnson-Laird, have ques-
tioned the status of these claims. Based on certain considerations of logic,
Anderson has claimed that, for every line of experimentation, one can
develop explanations in terms of either propositions or of imagery, and
there is no principled way to ascertain which is correct. Anderson con-
cludes that "barring decisive physiological data, it will not be possible to
establish whether an internal representation is pictorial or propositional"
(1978, p. 249). Johnson-Laird does not accept Anderson's proof but says
331
I I I I TowARD AN INTEGRA TED CoGNITIVE SciENCE
that it is unlikely that imagery disputes will be settled by psychological

experiment. In his view, a more productive discussion should revolve
around levels of representation. Although at one level a psychological
process may use only strings of symbols, at a higher level it may use
various sorts of representation, including arrays, dot matrices, and the like.
For various purposes, including the solving of certain kinds of problems,
it may be reasonable for humans to use mental images or, as Johnson-Laird
prefers to call them, "mental models" (Johnson-Laird 1983).
Pylyshyn s Penetrating Case against Imagery

Other commentators, however, argue that Kosslyn's claims are simply
wrong. Probably his most tenacious and probing opponent is the Canadian
computer scientist and psychologist Zenon Pylyshyn who has published
approximately a dozen major articles, all designed to refute the claim that
imagery merits consideration as a separate form of mental representation.
Like a persistent lawyer, Pylyshyn attacks Kosslyn at every level of analy-
sis and for every aspect of his studies (Pylyshyn 1979, 1981, 1984). But the
nub of his critique has to do with the tenability of Kosslyn's claim that
human beings can think-can solve problems or reason-by making use
of a separate medium called imagery. If there were such a biological func-
tion as imagery, it would be a fixed human capacity: in fact, individuals
would have to use the medium of imagery in order to solve certain kinds
of problem. But, in Pylyshyn's view, imagery is simply a product of sym-
bolically encoded rules and propositions, like beliefs and goals. In other
words, an individual has available lots of knowledge encoded in proposi-
tions and simply draws on these propositions in order to construct what
appears on phenomenal evidence to be an image (Pylyshyn 1981, 1984).
Pylyshyn's complex argument is worth following because it repre-
sents widespread views of how to conduct cognitive science and how to
conceptualize the mind. He begins with the flat assertion that cognition
is computation: the computer is not just a metaphor for the brain, but the
mind computes in a literal fashion. He goes on to distinguish two funda-
mentally different explanations for the behavior of a system (1984, p. 210).
The first appeals to intrinsic properties of the system, to processes that reflect
the operation of natural laws. (This is the level J. J. Gibson was actually
exploring.) The second explanatory form appeals to external properties-to
properties of the external world which the system must be able somehow
to represent. In this latter case, the explanatory principles have to take into
account the goals and beliefs of the system. This second form, which
touches on inferences rather than on reactions, particularly engages the
cognitivist. Here we enter the land of Chomsky and Fodor.
332
In explicating intrinsic properties, Pylyshyn has coined the termfimc-
tional architecture to refer to basic information-processing mechanisms of the
system for which a nonrepresentational account suffices. The functional
architecture includes the basic operations permitted by the biological sys-
tem, as well as such built-in constraints as the size of the memory, the
capacity of a buffer, the list of permissible operations, and the like. The
functional architecture cannot vary in ways that demand a cognitive expla-
nation (in terms of goals, beliefs, representations, and the like). Rather,
differences in cognitive phenomena can themselves be explained by ap-
pealing to the different arrangements that can obtain among the fixed set
of operations.
The notion of functional architecture allows Pylyshyn to cleave apart
two fundamentally different types of processes: those whose explanations
require appeal to certain kinds of representations, and those whose expla-
nations do not. Those processes requiring appeal to the representational
level are termed cognitively penetrable: they can be (and routinely are) affected
by an individual's symbolic processes, including beliefs, wishes, and the
like. Countenancing inferences, they fall outside the functional architec-
ture. In contrast are those capacities that are cognitively impenetrable: part of
the functional architecture, these processes are carried out in an automatic
and encapsulated way, impervious to an individual's particular beliefs.
Impenetrable capacities can be likened to the hardware built into a com-
puter; penetrable capacities are programmable, hence subject to change.
It would be desirable if the human cognitive repertoire were largely
impenetrable: then one could explain many processes as occurring neces-
sarily, courtesy of neural wiring. It turns out, however, that much of
human cognition is penetrable. And, in particular, in Pylyshyn's view,
imagery turns out to be cognitively penetrable: rather than being a requirement
of a certain kind of cognitive system (a part of the hardware), it can be
changed in all kinds of ways simply by one's giving (propositional) in-
structions to oneself. If the results reported by Kosslyn were supposed to
contribute to our knowledge of how the mind works, these functions must
be cognitively impenetrable: they must always apply when imagery is
putatively involved. But one can, in fact, claims Pylyshyn, change such
functions: they are cognitively penetrable. Put concretely, you can make
your attention jump from place to place in an image as easily as you can
make your scanning speed change--evidence that these activities are at the
behest of one's ideas and beliefs.
Pylyshyn revisits Kosslyn's map-location task but interprets it in a
different vein. Where Kosslyn believes that a subject is focusing on one
point after another in a mental image, Pylyshyn believes that the subject
is simply imagining the "real life" situation and how one would behave
333
in it. Thus, the experiment is probing tacit knowledge of the depicted

situation rather than perception of (or within) an imaginal medium. More
generally, the fact that information in an image is always labeled suggests
that the representation must have been interpreted before it was generated.
In Pylyshyn's view, the surface display dominating Kosslyn's theory
serves little function; its elements and its interpretation are at the behest
of a subject's goals, beliefs, and efforts. There are no intrinsic proper-
ties of the surface display: all are interpreted and subject to alteration
(Pylyshyn 1979, p. 562).
Pylyshyn concludes that it is best to think in terms of one set of
processes that are biologically determined-built into the hardware, so to
speak-and of another set of abstract symbols or propositions that can
simply be manipulated by a set of rules in a computer program. Images are
not built into the hardware and thus ought to be considered as epi-
phenomena secondary to the manipulation of symbolic elements.
But Kosslyn will not be drawn into Pylyshyn's particular distinctions.
He prefers to avoid reference to underlying abstract propositions and to
refer instead to various forms of representation-such as verbal and imagi-
nal varieties. Instead of the possibly misguided search for an irreducible
form of representation, Kosslyn finds it preferable to posit the internal
representations most effective in accounting for empirical findings. In fact,
the issue is not whether images may be derived from more primitive
propositional or symbolic representations, but rather whether a quasi-
pictorial image has emergent properties permitting its treatment as a dis-
tinctive form of representation. Efforts to explain all of the imagery results
reported earlier in terms of a propositional or symbolic code turn out to
be clumsy and roundabout, whereas the kind of model developed by
Kosslyn handles the current results neatly and also predicts suggestive new
ones. Pylyshyn may believe that his theory is more parsimonious because
he posits only one mode of mental representation, but Kosslyn's theory
can readily handle a whole range of data which Pylyshyn is unable to
address in a satisfying manner.
In this extended debate, Pylyshyn represents the mainstream of com-
puter scientists who have a long-standing commitment to a digital symbol-
computing model, where operations upon strings of symbols are the way
in which information is processed. He places a great deal of stock in this
symbol-centered approach:
The notion of a discrete atomic symbol is the basis of all formal understand-
ing. Indeed, it is the basis of all systems of thought, expression or calculation for
which a notation is available .... No one has succeeded in defining any other type
of atom from which formal understanding can be derived. Small wonder, then, that
334
many of us are reluctant to dispense with this foundation in cognitive psychology
under frequent exhortations to accept symbols with such varied intrinsic properties
as continuous or analogue properties. (1984, p. 51)
Pylyshyn finds a disturbing circularity in the claims for imagery. In

his view, a theory of cognitive processes should explain how imagery
comes to have its putative powers. One would need to show that there is
an imaginal mode of processing which is indeed an inviolable property of
cognition (having to do with the structure of the brain) and has to be
employed in order to answer questions about locations and distances and
the like. But, as Pylyshyn sees it, Kosslyn simply assumes a medium of
imagery and then purports to explain its character by indicating the ways
in which imagery works-the kinds of operation in which images are
involved. Pylyshyn concludes that the question about Kosslyn's work is
"whether we are entitled to take certain specific empirical regularities as
revealing general constraints arising from the nature of the ... representa-
tional medium or format, or whether such regularities are the consequences
of what the subjects believe, what they take their goals to be" (1980b,
p. 162).
For his part, Kosslyn concedes that there is a basic propositional level
of coding, from which images may well be generated, at least in part. He
takes his "anti-reductionist" or "anti-monist" stand on the legitimacy and
importance of a separate mode of representation which can be productive
in its own right. Not only do mental images play a causal role in thinking
but, once formed, they remain encoded in a form that can then be put to
various uses. Pylyshyn's notions are simply not productive: for example,
despite his criteria of cognitive impenetrability, he is not able to give
examples of any process (beyond the most elementary perceptual ones)
that is insensitive to beliefs, wishes, and other intentional phenomena.
Kosslyn has the weight of experimental evidence on his side, for when he
makes claims about imagery, or offers a simulation, he can invoke a great
deal of supportive data. In fact, as I have noted, neural evidence can now
be cited in support of Kosslyn's model. Of course, since Pylyshyn is trying
to make a basic conceptual point, he is not likely to be persuaded by yet
another study or empirical demonstration.
To my own way of thinking, Kosslyn has the stronger line of argu-
ment within the empirical cognitive sciences. If one is trying to model the
way the mind works, and a certain line of modeling consistently produces
rich and revealing results, then it is folly to dismiss that line just because
of some theoretical objections-which would be very hard to disprove.
(Indeed, in my own view, further research might well reveal several imag-
ery systems, analogous to the one described for the visual system, but
335
capturing and transforming other kinds of information, such as linguistic,

musical, or tactile.) Moreover, as Ned Block (1983) has pointed out, the
various questions anti-imagists have raised about the difficulty of finding
images (or "quasi-pictures") in the mind can actually be leveled against the
propositionalists as well. After all, it is hardly obvious what it would mean
for there to be propositions in the mind; it is just that the idea is somewhat
less vivid (somewhat less imagistic!) and may therefore be somewhat easier
to swallow.
But if Kosslyn's approach seems reasonable enough within the main-
stream of cognitive science, it has raised considerable hackles within the
wider philosophical community. Philosophers have raised the problem of
what it means when one begins to use the term imagery and falls readily
(and nonreflectively) into the use of various associated expressions: scan-
ning the image, focusing on particulars thereof, transforming the image,
and the like. Since it is clear that there are no pictures (or even any
"quasi-pictures") in the mind (who or what is there to see them? what is
the medium of the picture? and so on), the question must be raised just
what is gained, and what is risked, by elaborating on this metaphor
(Schwartz 1981). Georges Rey (1981) poses some unsettling questions. He
asks of images, which appear to exist in two dimensions, in what kind of
space do they live? Is the mental image really as brilliantly colored as the
"live" physical configuration? He also challenges Kosslyn's seductive
model. Cathode-ray tubes are perceived by human eyes, but surely there
is no further eye located in the recesses of the brain. People often report
that they see things that do not really exist-perhaps on this score, what
one should try to do is to explain what people report, including their recital
of images, rather than try to study images that may not exist in any
meaningful sense.
A Wittgensfeinian Critique
Pushing this line of criticism yet further is William Shebar (1979), a
philosophy student who worked with Kosslyn and then subjected some of
his ideas to a Wittgensteinian critique. As I noted in chapter 4, Wittgen-
stein had little regard for psychological theorizing: in his view, concepts
were often adopted unwittingly from certain seductive features of our
language. Thus, a Wittgensteinian approach calls for careful attention to
the language involved in such expressions as "imagining" or "having an
image," and is skeptical about images as an explanatory construct within
psychology.
As Shebar puts it, a careful look at the language involved in talking
about images reveals the many traps concealed therein. One should not
336
think of an image as a thing, for what kind of a thing is it? It is preferable
instead to talk about the action of imagining, which is something that
individuals legitimately do. You can tell someone to imagine a house-
because that is something that people can do-but you cannot tell someone
to "see a house," for seeing is a different kind of action from imagining,
and not one subject to voluntary control. By the same token, it is equally
misleading to analogize an image to a picture. One needs materials to make
a picture but not to make an image; yet this form of locution leads us to
think of imagery as a private slide show which we can see on the private
screens of our minds. Thus, you obviously cannot draw someone else's
picture, but you do have the capacity to imagine it.
Shebar objects to Kosslyn's characterizations of his results. The same
results can be explained without the need for an analogy with imagery.
Anyone knows that it takes longer to scan further distances on real pictures
and thus, when asked to imagine this process, conceives of it along the
same lines. The fact that it takes a longer time to press the button does not
imply the existence of a particular process called imagery-or a particular
element called an image-which is longer; it is simply an affirmation that
whatever happens when one scans a longer distance takes more time.
In Shebar's view, the term infernal representation is loaded from the start
with connotations of external representations on which it is modeled. But
it is not proper to talk about a hidden picture as a representation-for to
whom would it represent anything if one could see it? Shebar senses a
pernicious circle. Subjects are understanding the instructions given by
Kosslyn by grasping the very analogy that has been used to make the
predictions and is later used as an explanation for the result. As Shebar
puts it, a subject's understanding of the "inner process" talk in the instruc-
tions depends upon one's grasping of the "outer" process talk on which
it is modeled; and thus, one is bound to produce just those regularities that
look like evidence for the theory.
Shebar's critique is potentially devastating for imagery studies and
possibly as well for other areas of psychology which dabble in internal
representation. From his perspective, psychologists think they are study-
ing processes but are actually examining the effects of adapting a certain
way of talking about things. The researchers are using a form of locution
that actually reflects pretheoretical ways of discussing experience, rather
than mechanisms that can actually be investigated and understood. Ac-
cording to this account, the problems of psychology will be solved not by
newer data or more precise terminology but, rather, by the realization that
they are not genuine problems.
In my view Shebar's critique is tonic, perhaps slightly toxic, but not
as fatal as he might wish. There are, in fact, considerable risks in adapting
337
a certain vocabulary, for it carries associations that may be unconscious-

and, therefore, even more destructive to the users. But the cure for this
problem is more careful attention to one's way of talking, more scrupu-
lousness of one's theory, rather than necessarily abandoning a line of
study. Certainly, studying how individuals perceive is worthwhile-as not
even Wittgenstein would have denied; and if the appearance of internal
images seems to be part of perception, or seems to have properties akin to
"live" visual perception, such phenomena are worth studying also. But
since those capacities are less overt, less readily subjected to intersubjective
reliability, one needs to be all the more alert.
A Wittgensteinian perspective raises profound questions for cognitive
science. This aspiring discipline rests on the assumption that it is valid to
speak about internal representation: that is, about a separate level between
the nerve cells of physiology and the behavioral norms of a community.
Wittgenstein was unwilling to embrace this assumption, though he
avoided simple-minded behaviorist critiques of "internal constructs" and
introduced a sophisticated view of how the community provides ways of
conceptualizing the world.
It might seem that studies of imagery are made to order for a Wittgen-
steinian critique. It is relatively straightforward to talk about an individ-
ual's perceptions of objects or about one's actions in the world; and even
talk about language or music is directed toward a symbolic system that can
be readily analyzed. But discussion of those inherently mysterious
phenomena called mental images is bound to stimulate metaphors (like the
"mind's eye") and exotic models (like a cathode-ray tube) that will differ
in certain telltale respects from the phenomena being modeled. Food for
philosophical lions!
Cautious cognitive scientists, such as Pylyshyn, might seem well ad-
vised to jettison talk of imagery and to retreat to firmer grounds. There is,
however, another, bolder scientific ploy to which Kosslyn, Shepard, and
their colleagues can lay claim. If the case for cognitive science can be made
with respect to the relatively tenuous (and tendentious) phenomenon of
imagery, if the Wittgensteinian critique can be met there, then the case can
be readily transported to firmer ground. Viewed in this perspective, the
obstacles involved in studying imagery may actually confer considerable
power upon efforts undertaken successfully in this area.
Still, there is at least one sense in which Wittgensteinian skepticism
could prevail. Should a persuasive case be made for a tight cognitive-
scientific account of veridical perception and also for that second-order
form of perception involved in mental imagery, these forms of mental
activity can still be carried out wholly-or at least in large part-without
reference to language and to organizing conceptual systems. (After all, it
338
may well be that nonlinguistic animals have images; and there is little
doubt that they perceive in ways similar to human beings.) But any cogni-
tive science worth its name will eventually have to offer accounts as well
of those cognitive capacities that are heavily infiltrated or penetrated by
linguistic and conceptual processes. In entering this area, cognitive scien-
tists confront the most vexing issues within their science-and also those
issues about which human beings care most deeply.
339
12
A World Categorized
In defining the human being as a featherless biped, the Greeks were not
simply providing a succinct definition of their own species. They were
exemplifying their belief that common objects of the world can be clas-
sified into groups, and that these groups can be defined by certain criteria!
attributes. Thus, all human beings belong to a single class called humans,
and that class can be defined reliably by the whimsical phrase to which I
have alluded.
Consider the set of accomplishments inhering in this terse formula-
tion. While all living creatures respond in similar ways to entities they
deem similar, these responses are typically unreflective. Noting that
humans everywhere group entities together constitutes an important
achievement. Realization that these categorizations can form a nested hier-
archy is a further insight. For instance, a person treats different views of
a terrier as instances of the same terrier; groups together all terriers; then
groups terriers with poodles into the categories of dogs, dogs and cats into
the category of domestic animals, and on up the hierarchy to mammals,
vertebrates, animals, living things, and, simply, entities (Keil1979). More-
over, in addition to forming and organizing categories, human beings also
ferret out the traits defining each category. If one can supply a satisfactory
definition of a terrier, a dog, a domestic animal, and the like, one will then
have a rationale for placing every entity into one or more appropriate
categories.
From the days of Aristotle, such practices of naming, defining, and
categorizing have undergone philosophical scrutiny. By the middle of this
century, a certain position had become entrenched as the "right way" to
think about categories, concepts, and classifications (a trio of terms I shall
use here interchangeably). And yet in the past thirty-five years, during the
very period when cognitive science has been in the ascendancy, this view
340
A World Categorized
of how we categorize the world has undergone the most severe attack, until
today virtually no one holds it in its pure form.
In this chapter, I shall first review the major points of the classical view
of concepts and then consider the major attack on this position in the fields
of psychology, anthropology, and philosophy. Today it is no exaggeration
to say that the classical view of concepts has been replaced by a natural view
of concepts. The new view is not without its critics, including a fair num-
ber of commentators who yearn for the less ambiguous days of the classical
view, as well as radical scholars who doubt altogether the utility of invent-
ing categorization. After surveying a range of responses to the natural
view, I shall take stock of what has now been established about human
classificatory practices, and how this knowledge can inform an emerging
cognitive science.
The Classical View of Classification
I provided an initial glimpse of the classical view in introducing Jerome

Bruner's study of how individuals learn to form or attain concepts (Bruner,
Goodnow, and Austin 1956). Recall (p. 93) that subjects were asked to
recognize instances of geometric concepts like large red triangle or fall blue
cylinder. In such cases, a category was arbitrarily defined (any set of attri-
butes could have been targeted), and each item unambiguously fitted (or
failed to fit) into that category. The traditional recipe: a category and a set
of defining features, just like the featherless biped. Nor was this point of
view restricted to experimental cognitive psychology. Philosophers' favor-
ite definitions of concepts adhered to the same procedure: a bachelor is an
unmarried adult male. Anthropologists looked for similar structures in
their studies of kinship (an uncle is male, of an older generation, and not
a direct lineal antecedent). Even in the area of neuroscience, a search was
on for detectors that registered unambiguously to all lines that were ori-
ented in a certain direction but to none otherwise oriented.
Playing a classicist to the classical theory, I shall lay out its defining
features (see Smith and Medin 1981):
1. Categories are arbitrary. Nothing in the world or in our nervous
system determines how we must slice up our observations. Cultures and
languages do this work. Items can be grouped together in any number of
ways to form categories, and people can learn to identify or construct those
categories defined by their culture.
2. Categories have defining or critical attributes. All members of a
341
category share these defining attributes, no nonmembers share them, and

there is no overlap between members and nonmembers.
3. The intension (or set of attributes) determines the extension of a cate-
gory (which items are members). Hence, it makes no sense to talk about
a category as having an internal structure, with some items standing out
as better members than other items. Either the triangle is tall and red, or
it is not. Boundaries are sharp and not fuzzy.
These assumptions have been stated in bold form: one can find objec-
tions to such a sober vision of categories in writings dating back to the
British empiricists. Nonetheless, some variant of this classical position was
widely held until the middle of this century, when Ludwig Wittgenstein
and his followers hurled a challenge which has in recent decades been
reinforced by considerable behavioral-scientific work. Eleanor Rosch, pos-
sibly the cognitivist whose critique did most to undermine the classical
view, described the pervasive wisdom around 1950:
The processor was assumed to be rational, and attention was directed to the
logical nature of problem-solving strategies. The "mature western mind" was
presumed to be one that, in abstracting knowledge from the idiosyncracies of
particular everyday experience, employed Aristologian laws of logic. When applied
to categories, this meant that to know a category was to have an abstracted clear-
cut, necessary, and sufficient criteria for category membership. If other thought
processes, such as imagery, ostensive definition, reasoning by analogy to particular
instances, or the use of metaphors were considered at all, they were usually rele-
gated to lesser beings such as women, children, primitive people, or even to nonhu-
mans. (Rosch and Lloyd 1978, p. 2)
In one sense, this classical view of categorization was consonant with the
aura of cognitive science-a science built upon the unambiguity of the
computer. But just as the general view of the computer as a single all-
purpose logical machine eventually gave way to a plurality of computa-
tional stances, so, too, this general view of classification was not able to
withstand the force of arguments and empirical data from several quarters.
A strong blow against it came from work in a domain the classical view
had initially adopted as its own-the area of color naming.
The Universe of Color Terms
From a purely physical point of view, there are no indices to designate

where one color ends and another begins: the color spectrum (or sphere)
is a continuum throughout. Yet every human group has some means of
342
A World Categorized
naming and labeling colors, and these means of subdividing the spectrum
seem quite diverse. Some cultures have just two or three names for describ-
ing all colors; while other societies, including our own, feature a family of
terms-unmodified (blue) as well as highly modified (light sky blue). Since
each culture apparently carves up the color sphere as it sees fit, there are
apparently no natural laws at work here. Rather, one culture decides to cut
the color spectrum at a certain point (for example, at wavelength r ); while
another culture divides it at wavelength y, or divides it a different number
of times. The task of a person within a culture is to learn the color name
that that culture has arbitrarily hit upon, just as one learns the name of
family members, flora and fauna, and various human artifacts, ranging
from tools to forms of government.
One scholar who had been thinking along these lines was Roger
Brown, psycholinguist at Harvard University. Working in the early 1950s
with his graduate student Eric Lenneberg, Brown had become interested
in the way in which a culture's particular parsing of the color spectrum
may affect how individuals from that culture classify and later recall spe-
cific hues. In a review written in 1975, Brown has caught the Zeitgeist of that
period: he has also suggested the ways in which it came to be undermined
by Eleanor Rosch, who happens to have been his student nearly twenty
years later.
As Brown recalls in his review, he and Eric Lenneberg (1954) wanted
to test the Whorf-Sapir hypothesis: that is, one's conceptualization of the
world reflects the particular terms and concepts of one's culture (see also
chapter 8, p. 235). These researchers selected color for two principal rea-
sons: first, because colors can be described in an objective "culture-free"
way, drawing upon an extensive psychophysical apparatus; second, be-
cause the cross-cultural records documented many cultures with disparate
color lexicons.
For various reasons, the original study by Brown and Lenneberg actu-
ally involved only native speakers of English. Subjects were shown some
twenty-four colors and asked to name them: those colors that were readily
named were called "codable." Another group of subjects was then briefly
exposed to a small set of colors and thereafter shown a large set of hues
and asked to indicate which they had seen before and which they had not.
The colors in the new set included ones that had been rated as codable and
ones that had not been so labeled. The results provided support for a weak
version of the Whorf-Sapir hypothesis: specifically, subjects were some-
what better able to recognize those chips that had been rated as highly
codable than those that had not. The authors judged that the existence of
names within the American lexicon had exerted an influence on the sub-
jects' behavior, by making those colors that were readily named easier to
recognize.
343
I I I I ToWARD AN INTEGRATED CoGNITIVE SciENCE
Rosch s Critique of Classical Views
In the decade or so following the original Brown and Lenneberg inves-

tigation, several follow-up studies were conducted. This research pro-
ceeded in various meandering directions but was no more conclusive than
the original Brown-Lenneberg study. There was modest support for the
Whorf-Sapir hypothesis, little conclusive evidence against it. But around
1970, Eleanor Rosch (then Eleanor Rosch Heider) had the opportunity to
visit the Dani in New Guinea. The Dani, a Stone Age people, have but two
color terms-mola for bright, warm hues, and mili for dark, cold ones.
Curious about the effects of such an unusually limited color vocabulary on
the Danis' behavior with various color tasks, Rosch exposed the Dani to
forty color chips, representing four brightnesses and ten hues. Each subject
was asked to name each of the chips; then subsequently in a recognition
condition, a subject was shown a test chip, asked to wait in the dark for
thirty seconds and then asked to pick, from the array of forty, the one he
had seen. On the naming task, the Dani clearly confirmed that they are
from a culture different from ours. At the far end of the color curves, the
Dani agreed among themselves about which colors were mili and which
were mola. But intermediate chips were not consensually validated because
individual subjects located the boundary between the two color terms at
diverse points.
On the recognition condition, however, the results were completely
unexpected. The Dani turned out to recognize colors in a manner very
similar to the Americans-making the same confusions, for example-
though their overall scores were not as high. Thus, differences in naming
structure were not paralleled by differences in the way in which colors
were stored in memory or accessed for recall (Heider 1972; Rosch 1973a;
Rosch 1973b).
As a further check, Rosch and her colleague Donald Olivier (1972)
selected hues immediately adjacent to one another in the color sphere and
matched on saturation and brightness. Sometimes the members of a hue
pair lay on the same side of a color name (whether English or Dani); at
other times, they lay on opposite sides of the line, hence reflecting different
color categories (for example, green and blue). Even with this simplified
task, recognition proved equally keen for perceptually adjacent hues
whether or not they lay on different sides of a language line-decisive
evidence against a classical Whorfian position. As Brown puts it: "The
fascinating irony of this research is that it began in a spirit of strong
relativism and linguistic determinism and has now come to a position of
cultural universalism and linguistic insignificance" (1975, p. 152). Or, in
Rosch's own words, "In short, far from being a domain well suited to the
344
A World Categorized
study of the effects of language on thought, the color space would seem
to be a prime example of the influence of underlying perceptual cognitive
factors on the formation and reference of linguistic categories" (Heider
1972, p. 20).
Why, then, was this set of findings damaging to the classical view of
concept formation? For many years, it had been assumed that the lines
between colors were arbitrary, drawn as seen fit by a culture; individuals
would simply mirror these boundaries in their own classificatory and mne-
monic behavior. Now Rosch was calling into question these various lines
of argument. The naming practices turned out to be incidental. The ways
in which individuals from different cultures remember colors seemed to
reflect the organization of the nervous system, not the structure of particu-
lar lexicons. Certain colors are "good" instances of a color because of the
physiology of the human visual system and not because of specific naming
practices. Indeed, a lexicon codes aspects of color that are already salient
rather than making these aspects salient.
Rosch was by no means content simply to conduct studies of color
naming among an exotic population (Mervis and Rosch 1981; Rosch 1977,
1978). Following her return to America, she probed a wide gamut of
domains and concluded that the story on color is highly pertinent else-
where as well. According to the classical view of classification, based
primarily on the use of artificial stimuli, each category is defined by a finite
list of criteria! features: members display these features; nonmembers do
not. But this classical picture does not apply to the world of natural objects
-like birds-nor does it prove particularly illuminating in the realm of
numerous man-made objects, like furniture or tools. In the world of every-
day reality, highly correlated (non-independent) features prove the rule.
For example, given the capacity to perceive feathers, furs, and wings, a
perceiver soon comes to realize that in the empirical world, wings occur
with feathers more than they do with furs. There is, in other words,
considerable redundancy in the appearance of members of the same cate-
gory-not the independence of features posited in the classical view. Rec-
ognition mechanisms exploit these redundancies.
There are other intrusions from the world as it actually is. For exam-
ple, with respect to man-made entities, objects with one sort of appearance
are more likely to be grouped together as chairs (because they possess the
potential for being sat upon) while another group of objects is more likely
to be grouped as drinking vessels (because they possess the potential for
being held and poured from). While some characteristics may appear to
demark chairs and others to demark drinking vessels, one looks in vain for
a set of defining criteria; instead, the kinds of action these objects seem to
elicit (or afford) constitute a more useful aid to classification.
345
II I I TOWARD AN INTEGRATED COGNITIVE SCIENCE
Rosch has discerned a basic structure which seems observable across

a wide range of categories. Categories are built around a central member
or prototype-a representative example of that class which shares the most
features with other members of the category while sharing few, if any,
features with elements drawn from outside the class. A robin is a more
prototypical bird than a chicken or a penguin: accordingly, it is more
readily recognized as a bird and less likely to be misclassified as a member
of another category (such as a mammal or a fish). Similarly, a sedan
emerges as a more prototypical car than does a convertible or a limousine.
Indeed, even with respect to artificially designed categories-for example,
a series of dot patterns or a set of nonsense figures whose members have
been fashioned according to certain prescribed criteria-the same findings
apply. That is, the data on recognition or memory of these forms seem
more adequately accounted for by assuming that subjects are erecting
prototypes, than by assuming that they are looking for (or devising) a fixed
list of features. The classical view founders even on its own home turf.
Rosch has called attention to other aspects of category structure as
well. In many common categories of experience, she discerns a certain level
at which subjects most readily can learn names, have ready access to them,
remember them, and the like. She dubs this the basic level. Objects at the
basic level within a category share many perceptual similarities and func-
tional characteristics. For example, in the realm of furniture, a chair is a
basic-level object; with respect to the animal world, a dog or a bird would
be a basic-level object. The basic-level object contrasts with a higher level
called the "superordinate" (furniture, in the case of a chair; animal, in the case
of a bird), and can also be contrasted with the "subordinate" level (rocking
chair in the case of chairs; robins or wrens in the case of birds). Young
children strongly favor naming all objects at the basic level-all four-
legged beasts thus being "dogs" for a time, and the categories of animal
and collie emerging only later. Ultimately individuals become able to name
and classify at these various levels but tend, whenever possible, to embrace
the basic level of organization.
As a final point, Rosch questions the existence of strong or fixed
boundaries between categories. While one can, of course, construct catego-
ries that have defining features and fixed boundaries, categories in the real
world tend to have fuzzy boundaries and to blend into one another. These
views echo those introduced some years earlier by Ludwig Wittgenstein.
In propounding such ideas, Rosch challenged each of the major tenets
of the classical view. Rather than being arbitrary, categories are seen as
motivated. Conceived of in a Gibsonian vein, categories reflect the percep-
tual structure of the perceiver, the kinds of actions one can carry out, the
physical structure of the world. Categories do not have criteria! features
346
A World Categorized
but harbor prototypes, with less prototypical members being apprehended
with reference to the extent that they resemble (or fail to resemble) the
prototype. Categories exhibit internal structure and that structure yields
psychological consequences: basic-level objects tum out to be the most
readily named and recalled; less psychologically accessible, but still useful
and necessary for various purposes, are the finer-grained level of the sub-
ordinate and the overarching level of the superordinate. Finally, thecate-
gories do not have firm boundaries: many members sit astride the border
of two or more categories.
Several factors may have contributed to this state of affairs in the
world of categorization. Having evolved over many years to be able to deal
efficiently with their environment, human beings tend to group together
into categories entities that appear similar to their perceptual apparatus or
call upon similar actions, or both. Moreover, what looks similar is not in
the least arbitrary. For example, our visual system is so designed as to treat
certain reds as being better than others, and to draw the line between red
and orange, or between orange and yellow at specific points on what looks
(to the instruments of the physical scientist) like a continuous spectrum.
This view makes clear contact with what is known about the physiology
of color vision, and links are forged as well with recent work in logic: the
new variety of fuzzy set logic, which deals with degree of membership in a
class, turns out 1:o fit well with the claims being made by Rosch and her
colleagues. And so one telltale finding about color naming in a remote
Stone Age tribe has fostered a rev9lution in the way we conceptualize
concepts.
While nearly ~veryone! agrees that the Rosch findings undermine a
strong view of t~e classical theory of concepts, critics question whether the
natural view, taken alon~, can replace it. In one study, Sharon Armstrong,
Ula Gleitman, and Henry Gleitman (1983) examined the structure of a
category that was clearly defined in the classical criterial way: the category
of odd number. To their delight, these authors determined that individuals
are as likely to organize such a concept around a prototype as they will a
natural category like bird or vegetable (seven turns out to be the prototypi-
cal odd number). The researchers concluded that Rosch has no way of
distinguishing between a classically defined and a natural concept. How-
ever, it seems to me that this line of evidence could as readily be cited as
showing that even "true" classical concepts have the kind of categorical
structure Rosch has ferreted out elsewhere. According to the latter analy-
sis, the tendency to discern naturalistic features in all categories is viewed
as a confirmation, rather than a refutation, of the natural view.
Another critical view, put forth by Daniel Osherson and Edward
Smith (1981), asserts that the prototype theory cannot account for the
347
I II I TowARD AN INTEGRATED CoGNITIVE SciENCE
ways in which more complex concepts (for example, the concept pet
fish) are apparently composed of elementary concepts. These authors pro-
pose a hybrid theory: that is, while categorization continues to need a core
concept, along the line of classical concept theory, this core aspect must
be combined with an identification process (the way in which one deter-
mines category membership). Rosch's approach is seen as an explanation
of how one identifies instances of the natural and artificial sort. In the view
of Osherson and Smith, such a hybrid theory is important because "the
ability to construct thoughts and complex concepts out of some basic stock
of concepts seems to lie near the heart of human mentation" (1981, p. 55).
To my mind, neither the findings of Armstrong and colleagues nor the
Osherson-Smith argument undermines the broad thrust of the Rosch posi-
tion. The classical theory remains feeble; the natural view emerges as a
more veridical description of how individuals form and utilize categories.
And yet the two lines of criticisms that I have reviewed do make an
important point: certain aspects of human cognition-such as the capacity
to apply the definition of odd number in a reliable manner or the ability to
form more elaborate concepts out of simpler ones-seem inexplicable in
terms of the natural view. At least some aspects of categorization may
employ the kind of computational operations classicists cherish.
Berlin and Kay on Basic Color Terminology

Remaining within the field of choice, Brent Berlin and Paul Kay of the
University of California at Berkeley provided intriguing evidence about
the "naturalness" of color names (1969). As I have noted earlier, these
anthropologists studied the color-naming practices of individuals drawn
from twenty genetically diverse languages, ranging from Arabic to lbo
(Nigeria) to Thai, and found that cultures differ dramatically from one
another in the number of basic color terms they possess, from two (like the
Dani) to eleven (like the English).
Dwarfing these differences are two important phenomena, both docu-
mented by Berlin and Kay. First of all, when confronted with chips span-
ning the full spectrum of hues, their informants selected the same focal
areas for colors, irrespective of whether they had names for them: that is,
informants from these diverse cultures agreed about what was a "good
blue" or a "poor green," even if their culture lacked names for these colors.
Indeed, the differences among individual speakers of one language were
comparable to those among speakers of different languages. Berlin and Kay
attribute this universal agreement to the structure of the nervous system
of the primate, which renders certain hues more salient than others (de
Valois and Jacobs 1968).
The second and remarkable finding was a fixed order in the construe-
348
A World Categorized
tion of color lexicons. Studying color lexicons from ninety-eight different
cultures, Berlin and Kay uncovered the following sequences. If (like the
Dani) a culture has only two color terms, terms will code for black and
white. If a third term is added, it will be red; fourth and fifth will be yellow
or green (followed by whichever one of the two has not been used); blue
and brown will be the next pair, competing for sixth and seventh places;
and purple, pink, orange, and gray will be the last four names to be coined.
These results are virtually impossible to account for through a series of
historical accidents or diffusions. Instead, the kinds of distinction cultures
make among colors, and elect to capture in their lexicons, apparently reflect
distinctions most salient in their perception of the world, presumably
(again) for neurophysiological reasons.
The Berlin and Kay findings were as epoch-making in the area of
anthropology as Rosch's were in psychology. Just as psychologists had to
reconsider their beliefs in classical definitions of concepts and in the arbi-
trariness of categories, anthropologists had to re-examine their beliefs in
the flexibility of naming and categorization schemes and in the influence
of language on thought structures. But some anthropologists feel that the
conclusions drawn by Berlin and Kay are too broad. John Bousfield, for
example, suggests that it is one thing to know that the color spectrum is
divided into two segments, but quite another to translate the terms as
"black" and "white." In his view, such a pair of terms would acquire far
more meaning (since they are the only ways to talk about colors) than they
do in a multicolor-term culture:
It is absolutely crucial to remember that we would be changing the meaning

of these two English terms if we now had to assign more or less every hue and
shade in the room to one category or the other. We could do so, just as the taste
of wines can be located on the dry-sweet dimension .... But we have lost some-
thing if we do not realize that this change has occurred in the meaning of our terms,
and if we think instead that their terms "mean the same" as ours. (1979, p. 208)
In his critique, Marshall Sahlins, an anthropologist at the University of

Chicago, goes even further (1976). Initially, he accepts the Berlin and Kay
finding without modification, saying that "relativism will simply have to
come to grips with the cross cultural regularities of color categorization"
(1976, p. 2). But he proceeds to interpret the findings in virtually the
opposite way. Rather than agreeing that the perception of colors directs a
culture's naming practices, Sahlins maintains that the human perceptual
apparatus itself is exploited by more pervasive cultural concerns: "The true
ethnographical existence of color terms and percepts [inheres in] their
actual cultural significance as codes of social, economic and ritual value"
(p. 8). In Sahlins' s "semantic" view, colors have nothing to do with a score of
sample chips-an arbitrary invention of Western technology-but, rather,
349
III I ToWARD AN INTEGRATED CoGNITIVE SciENCE
signify crucial cultural differences between life and death, noble and com-
mon, pure and impure. Even in our culture, terms like red or yellow, or blue or
green are appropriated to emphasize significant distinctions in "charged"
domains like politics, bodily states, religion. It is owing to their deep
concerns about phenomena like anger or patriotism or mourning-
phenomena that become associated with characteristic hues-that humans
attend to colors at all. To adopt this semantic point of departure, Sahlins
insists, is not to ignore biology but simply to assign it a proper place.
To my mind, it would be a mistake to get involved in a "chicken-egg"
question of which came first-the cultural distinctions motivating all hu-
mans or the ability to discriminate colors per se along certain lines. I read
Sahlins as making a different point. As he sees it, one does not confront
perception per se, except perhaps in an artificial situation where one asks
subjects to examine color chips. Perception is always marshaled in the
service of some cultural end. And so, a la Levi-Strauss, one ought to think
of the perceptual system-and globally, of the mind-as an implement of
culture, as an organizational structure to be exploited by the human cul-
tural enterprise. As Sahlins concludes, "It seems to me that Basic Color
Terms opens up very exciting prospects for an ethnography of color whose
general aim, quite beyond the determination of the empirical correlates of
semantic categories, might consist especially in the correlation of the semi-
otic and perceptual structures of color. For colors, too, are good to think
(with)" (1976, p. 16).
In such reactions to the Berlin and Kay findings, we see the same
impulse for territorial preservation which greeted the Rosch work. Just as
some psychologists felt the need to defend the classical view of concepts,
so, too, some anthropologists seek to ensure that the terrain of culture does
not disappear now that some potent universals have been demonstrated.
Perhaps scholars will eventually find some way of blending the strengths
of the two points of view, as Osherson and Smith attempt in psychology
and Sahlins attempts in anthropology. But to my mind, neither of these
lines of criticism undermines the main point of the natural view of catego-
ries. They are commentaries or cautionary notes, rather than convincing
revivals of the classical view in psychology or anthropology.
A New Philosophical Cast on Concepts
Just about the time that Eleanor Rosch had repaired to the wilds of New
Guinea, a shift in many ways parallel to the trends in psychology and
anthropology was coming to a head within the philosophical community.
350
A World Categorized
As has happened more than once, this philosophical shift was previewed
in the thinking of Ludwig Wittgenstein (1968). In the early part of the
century, as a principal contributor to, if not the inspirer of, the Vienna
circle (see pp. 62-65), Wittgenstein had stressed the importance of logic
and the need for precision in language (as well as the desirability of
keeping silent when such linguistic accuracy was not possible). But in the
second phase of his scholarly career, Wittgenstein came to focus on the
way in which ordinary language is used, and arrived at the controversial
conclusion that philosophical problems typically inhere in one's use of
language. He saw language as a loose and fragmentary set of elements and
as the necessary means of communication among individuals, but as prone
to obscure as well as to enlighten because it is the web through which all
other experience necessarily passes. Concepts are neither mental constructs
in the head nor abstract ideas in the world, but ought to be seen as
abilities which individuals can employ in ways acceptable to the rest of the
community-roughly speaking, as ways of accomplishing things.
In Wittgenstein's skeptical view, the most an analyst can hope to
achieve is greater insight into how language as a system works, and into
how our own ideas have come to be formed by the linguistic practices of
our community. Efforts to figure out what "really" happened, to sweep
aside the veil of language, are doomed to fail. And the glorification of logic,
or of abstract concepts bereft of utility within a community, is an irrele-
vant move, lacking philosophical force.
While Wittgenstein was calling for a radically different view of lan-
guage and conceptualization, the attack on logical empiricism was also
being pursued by once-sympathetic philosophers in the Anglo-American
tradition like W. V. 0. Quine (1953) and Nelson Goodman (1955). As I
noted in chapter 4, the distinction between the analytic and the synthetic,
the contrast between the immediate and the mediate, the possibility of
verifying observational sentences-these were all being spumed. More
modest goals for philosophy, or even the merging of philosophy with
psychology, as a kind of "naturalized epistemology," were being advanced.
Ultimately, a still more radical attack on the classical view of concepts
was put forth virtually in tandem by the American philosophers Saul
Kripke (1972) and Hilary Putnam (1975b ). This pair of scholars came to
question the belief that the world is a welter of sensation which can be
parsed with equal plausibility in an indefinite number of ways. Rather, in a
move analogous to the Roschian maneuver within psychology, these
philosophers embraced a view of naming and classifying that smacked of
the rankest realism. Kripke, Putnam, and their associates argued that there
is a real structure to the world, and that much of our conceptual armament is
designed to capture this genuine and attainable structure. Putnam (in his
talk of "natural kinds") and Kripke (in his talk of "rigid designators")
351
assumed that there are many objects, ranging from gold to goldfish, and also
many entities, like the persons "Richard Nixon" or "Greta Garbo," that
ought not be defined in terms of a list of criteria! attributes. For practical
purposes, these concepts have no definition, no intension; they have only an
extension, or a relation extending from the term to its concrete referent in
the world. Richard Nixon is the person whom his mother named as such; he
cannot be anyone else; and the only way to make sure that he is Richard
Nixon is to have a confirmed history that he is the same individual that his
parents so dubbed in the year 1913. By the same token, gold is what it is;
scientists may define it today as the atomic weight 79, but it will remain gold
even if the scientist's understanding of gold changes.
According to this view, all instances of the same natural kind possess
a definite underlying structural property in common. Thus, all lemons
ultimately possess the same structure even though any two randomly
selected lemons may lack a specific perceptible property in common. A
category is organized around stereotypes that enable a layperson to recog-
nize exemplars: such an organization proves essential inasmuch as one will
not necessarily be privy to the underlying genetic properties all lemons
really share.
As Stephen Schwartz (1979) has explained, we rely on experts to help
us to determine the appropriate use of such natural-kind terms. We assume
there is someone (or some procedure) that can confirm the identity of
Richard Nixon; we assume there is someone who can confirm that this
sphere is a lemon or that this metal is gold. Such natural-kind terms prove
susceptible to the formation of stable generalizations, such as, "If you do
X to gold, it will turn ...." To use the natural-kind term appropriately,
you do not need to know the trait governing the extension-indeed, at
present no one may know it; you just have to believe that it can be
discovered eventually. In other words, a nature is there (in principle) to be
discovered.
The question arises why we have natural-kind terms. Schwartz voices
the opinion that such terms turn up whenever the same stuff or thing
characteristically assumes a lot of different forms. Water can take different
forms, as can diseases, animals, plants, and so on. Seeing these changes
occurring on the "same" object, one assumes that there is some underlying
trait making the stuff continue to be of the same kind. In contrast, "artificial
kinds" do not necessarily exhibit dramatic changes: cars do not necessarily
rust, rings need not bend: these are simply vicissitudes to which any
man-made object may be subject. Schwartz goes on to discern a unity in the
changes exhibited by natural-kind terms, as all members of a particular kind
must go through the changes. Something of a law-governed nature must
remain the same in the transition from ice to water or from larva to butterfly.
352
A World Categorized
"Likewise the reason why I believe that tigers share an essence, the reason
why I am compelled to postulate an underlying trait for tigers that makes
tigers tigers, is that big tigers, by a perfectly natural, unconscious, and in-
eluctable process, produce little tigers" (1979, p. 315). Or to put it in Krip-
ke's terms, we come to think ofobjectsin termsofwhatcouldhappen to them,
in terms of the modal possibilities, the regularities and the exceptions.
To those somewhat removed from philosophical frays, the move to
natural-kind objects, the willingness to accept an ostensive definition,
("This is gold") or a history of occurrences ("This has always been
Nixon"), rather than a list of necessary and sufficient conditions ("Gold is
A,B,C, or Nixon is X, Y,Z) may seem innocuous enough. But within the
philosophical world, the moves first launched by Wittgenstein and then
extended in new and unanticipated directions by Kripke and Putnam are
considered very radical. Some sacred cows dating back to Aristotle are
directly challenged; some strong and biologically oriented contentions ("A
lemon has an essence ... ")are being introduced to a world that has been
comfortable in its empiricism and its nominalism. Rather than meanings
being located inside someone's head, the Putnam-Kripke tradition sug-
gests that meanings are located in the world-an imperative imposed by
the real structure of objects and the ways in which individuals come to
learn about them.
Discussion of the new theory of reference has been widespread within
the philosophical community. The talk of essences, of a "real" structure
in the world, of a connection between proper names (like Richard Nixon)
and natural-kind terms (like lemon or gold) is sometimes seen as a reac-
tionary move, one that recalls errors to which philosophers from a bygone
day were prey, but that had ostensibly been exorcised by several centuries
of empiricist philosophical analysis. The practice of collapsing diverse
elements into the same category because of ostensible underlying struc-
tural identities is seen as risky. Indeed, the reversion to realism has such
strong philosophical implications that even Putnam himself (though not
apparently Kripke) has had second thoughts about the dangerous lair into
which such an analytic line may lead.
Yet whatever misgivings an individual philosopher may feel, an ob-
server of the cognitive sciences is struck by the extent to which parallels
were occurring across several fields. The findings emerging empirically
from the laboratories of psychologists and from field trips to New Guinea
were closely reflecting the discussions engaged in by scholars who might
never think to conduct an experiment or join an anthropological team but
were deeply steeped in over two thousand years of argumentation on
related topics. From these widely different avenues came a common rejec-
tion of the classical view of concepts, a sharp impatience with criteriallists
353
I I I I TowARD AN INTEGRATED CoGNITivE SciENCE
of attributes, and a correlative belief that there are natural categories in the
world, best defined by examples (called variously "stereotypes" or "proto-
types"), and possibly reflecting underlying structures, which we as hu-
mans have the potential to recognize and to understand. In each case, the
extent to which the revisionist claims could take over the field remained
. in doubt, and some critics yearned for the good old classical days: yet the
challenge posed to the classical formulation was widely recognized and
frequently endorsed .
. · We have, then, a prototypic;al example of how several disciplines in the
cognitive sciences have combined to suggest a solution to a long-standing
philosophiCal puzzle. In this particular case, the puzzle surrounds the
human capacity to group together elements as members of a category, and
then to distinguish that category or class from others in the same general
domain of experience. A consensus had developed over the years that such
categorization was relatively arbitrary, with the culture determining which
entities ought to be grouped together. Most any set of criteria could be
proposed; and once proposed, these criteria could be applied reliably to
instances, as means of determining which belonged t0 the category and
which did not. Gasses were relatively clear; boundaries, relatively fixed.
Thus, it was appropriate for researchers like Jerome Bruner to conduct
experiments using artificially contrived classes: after all, they were thought
to represent well the kinds of classes we must utilize every day.
From the fields of psychology, anthropology, and philosophy, work-
ing initially in relative independence of one another, but possibly reflecting
the same factors or spirit, came an increasing realization that this classical
view is wrong. Categorization is anything but artificial and unambiguous;
rather, it relies on information in the natural world to which we as a species
are geared to respond. Categories have an internal structure, centered
around prototypes or stereotypes, with other instances being defined as
more or less peripheral depending upon the extent to which they share
pivotal features with the central prototype. Perceptual information is cru-
cial in defining the dimensions of a category; language, for the most part,
follows upon the discriminations made by the individuals, rather than
playing a controlling role in how one classifies in the world.
There is an irony in the decline of the classical theory-an irony we
have encountered in other contexts. The classical theory of concepts was
tailor-made to a computational model of mind-a set of precisely defined
dimensions which could, taken conjunctively, yield a category. Computers
were the perfect devices for simulating (or epitomizing) this mode of mind.
It is therefore striking that, at the very time that science had available a
device fully equal to simulating a certain view of concepts, that view was
found to be radically flawed. Not surprising, therefore, that there was
resistance across several disciplines to the relatively radical implications of
354
A World Categorized
the natural view of categorization. Many analysts hold out for a range of
categories, some of which are relatively natural, others of which are more
arbitrary and lend themselves readily to the kinds of analyses put forth in
the classical theory. And it is certainly true that the natural theory deals
better with colors or plants than with legal terms, religions, or self-con-
cepts; moreover the extent to which the natural theory handles man-made
artifacts satisfactorily is far from settled.
Indeed, to my mind, the principal challenge to the natural view of
categorization inheres in the extent to which it can move beyond an expla-
nation of those basic-level entities that we are "biologically prepared" to
perceive and classify. Color is most likely to lend itself to a natural-kind
account, because our perceptual system is primed to consider certain colors
as focal; and there is every reason to think that primates the world over
see colors in similar ways (unless they happen to be color-blind). The
evidence concerning other "natural" entities-such as plants, fruits, or
animals-is not as well worked out, but there is reason to expect that an
account in the natural-kind tradition may suffice (Berlin, Breedlove, and
Raven 1973).
But much of our classification in everyday experience extends beyond
the objects evolution has prepared us to classify in certain ways. There are
the multitude of man-made objects, ranging from tools to machines to
works of literature or art. There are more abstract concepts such as political
principles, religious precepts, belief systems, and economic laws. And there
are concepts that impinge much more upon an individual's own personal
concerns: concepts having to do with personality, motivation, sexuality,
emotions, and the self. Anthropologists in the ethnoscientific tradition have
attempted to apply to these concepts the approaches that worked in simpler
areas like kinship systems, plants, and colors. As I noted in chapter 8, these
attempts have not been notably successful; and, in fact, the discipline of
ethnoscience, in both its classical and its natural form, is quiescent at the
present time. One might say that the critique of the classical view has been
relatively successful but that the natural approach has not yet proved
genuinely illuminating with more complex, less "built-in" concepts.
Can Categorization Be Studied from

a Cognitivist Perspective?
Which leads to a more radical critique, of the sort associated with Clifford
Geertz of the Institute for Advanced Study, and other members of the
school of symbolic anthropology. Geertz (1973, 1983) opposes his position
355
to the central cognitive-scientific "dogmas," where thought is viewed as

unified, bound to the individual psyche, and governed by statable laws; in
place of this Newell-Simon vision, Geertz urges a view of thought as a
collective product, which is coded differently across cultures thanks to
historical forces that have exerted their effects over the millennia.
The existence of numerous cultures with diverse and idiosyncratic
histories makes comparison across cultures difficult, if not suspect. Rather
than comparing the psychological processes of one population with an-
other, one must ponder the commensurability of the conceptual structures
of one community with those of another community. Geertz sees little
reason to expect that a mimicry of the natural or the physical sciences will
be helpful here. Rather, he calls for the utilization of approaches associated
hitherto with the humanities. The anthropologist bent upon understand-
ing the conceptual structures or categorizing practices of an alien commu-
nity may share more with a literary critic trying to understand a text, with
a historian trying to make sense of documents, or with an art critic analyz-
ing a painting than with a chemist mixing elements or a biologist carrying
out experiments. In this approach, the analysis of culture is not an experi-
mental science in search of laws but is rather an interpretive or "her-
meneutic" scrutiny in search of meaning. The method needed here is an
imaginative leap where one tries to place oneself inside the head, or the
experience, of the "native." The preferred route is through careful idio-
graphic case studies: one studies the categories of religion or animal life by
observing the practices of daily living and not by requesting definitions or
administering tasks of classification.
Symbolic anthropology alerts us to another risk. Critical issues in the
study of perception, recognition, classification, logical reasoning, and other
cognitive operations may not be given to us by nature in the way, say, that
the movement of the heavenly bodies or the laws of the atom might
reasonably be taken as given, even by the proverbial observer from Mars.
Rather, our ways of conceptualizing precepts, concepts, and the like are the
products of a particular intellectual and cultural history, traceable in sig-
nificant measure to a mode of analysis that arose at the time of the Greeks.
We are involved in the pursuit of an understanding of concepts that come
out of our own philosophical and historical tradition; and we make a grave
error in assuming that this agenda has to be the agenda, the only feasible
one to pursue. Perhaps the Hindu or Dani tradition would parse cognition
in a radically different way or reject the concept of cognition altogether.
Such a line of argument makes contact with the critique of work on color
vision put forth by Sahlins (1976): on this account, the issues that in-
dividuals or cultures choose to make central dominate whatever cognitive
operations are provided by nature.
356
A World Categorized
Of course, resolution of so fundamental a controversy cannot be pro-
vided in advance: it is, as they say, an empirical issue or perhaps a question
of which line of explanation will prove most satisfactory and revealing in
light of the canons judged to be relevant-be they canons of the natural
sciences, the social sciences, the humanities, or the arts. It is not, however,
simply a matter of taste.
In my own view, even if the processes by which individuals reason
or classify may be similar the world over, the actual products and the ways
they are thought about may be so different as to make illuminating gener-
alizations elusive. And to the extent that particular concepts are the result
of biases lurking within the linguistic system of a given culture, one risks
entering a Wittgensteinian-Whorfian morass from which few investigators
have yet returned. It seems to me that the important lines to be drawn are
between those concepts and those principles of reasoning that have
evolved since the early hominid era, as against those that have developed
only comparatively recently and reflect the particular history, customs, and
values of one or another social or cultural group. Oearly, there are ways
in which humans form concepts of colors or of animals, and these may well
be the same the world over. In formulaic terms, concepts may not be
formed classically, but a naturalistic approach to certain aspects of concep-
tualization and reasoning appears possible.
But when one enters the world that is entirely man-made, in arts,
rituals, or sciences, the possibility for effective generalization seems less
evident. Here, forces of history and culture may prove so dominant that
the differences may outweigh the similarities. While certain generaliza-
tions can be formed, the ones that seem most revealing about a particular
circumstance may not extend beyond that circumstance. What tells us the
most about the Balinese may tell us the least about the Belgians, and vice
versa.
Scientists cherish their own reflections and use them to guide their
studies. They devise artificial, criterially defined categories (else their work
could not begin), and they follow strict canons of logic (otherwise their
work would not be accepted). In their efforts to model thought processes,
they have sought such categories and such operations in normal individu-
als. But in their search for marks of the scientist in nonscientific man (or
even in the scientist, when not "playing" scientist), these investigators
have been disappointed. Faced with an artificial concept, the subject at-
tempts (if inappropriately) to treat it as if it were a natural concept. Asked
to reason logically, subjects have embraced images and stereotypes. So we
confront the paradoxical situation where scientists have attempted-for
experimental purposes-to make the natural world artificial; but their
subjects, faced with the difficulty of dealing with these artificial concepts,
357
I I I I TowARD AN INTEGRA TED CoGNITIVE SciENCE
have simply (if sometimes inappropriately) invoked their naturalistic

modes of reasoning and classification.
Returning to the issues raised in the earliest philosophical debates, we
find that progress has indeed been made. At the very least, classical views
have been strongly challenged. Among natural-kind concepts, there are
some factors reminiscent of "ideal forms" and "universals"-Plato might
take some comfort in this vindication. Similarly, there is modest support
for the view that, under certain circumstances, the language one uses does
influence the way in which one conceptualizes the world; indeed, as one
moves toward complex, abstract, and less immediately perceptible realms,
the role of one's symbolic systems may become predominant. Yet in those
areas investigated thus far-like the classification of common objects or
common sensory experiences-the position of the naked nominalist or the
rabid Whorfian is not supported.
And how does cognitive science fare in this cooperative enterprise?
Once again, progress has depended upon convergences among disparate
fields. The particular disciplines most intensively involved in issues of
classification have been philosophy, where the problems have been
defined and discussed; psychology, where certain critical experiments have
been devised; and anthropology, where pertinent cross-cultural studies
have been carried out. Scholars such as Eleanor Rosch, Brent Berlin, Paul
Kay, and Hilary Putnam epitomize this ecumenical spirit. Until now, the
constraints on classifying imposed by neurophysiology have been modest;
still, when results like those with color naming prove consistent with what
is known about physiology, everyone feels on firmer ground. It is, in part,
the immense distance between our understanding of abstract concepts and
our understanding of the nervous system that renders remote, at the
present time, the hope for a conclusive account.
No less than was the case with perception and imagery, probing
critiques have been leveled at the whole enterprise. Just as Gibson has
questioned the representational views of perception, and as Wittgenstein
has questioned the explanatory power of imagery, so Geertz and other
anthropologists wonder whether comparative ethnoscientific studies of
categories valued in the West can illuminate the most vaunted forms of
thought. The anthropologists may well be correct in their view that com-
plex concepts elude current investigative techniques; and, at least for a
while, more of interest and importance may be learned through approaches
informed by literary sensibility. At the same time, the processes whereby
categorization takes place are recognizable across diverse populations and
diverse domains of experience; in that sense, we know more (and we know
differently) about categorization than we did before the advent of cogni-
tive science.
358
A World Categorized
In confronting questions of categorization, we have moved into a
distinctly human terrain, involving both language and reasoning capaci-
ties. Few would quarrel with the claim that this area is less advanced than
the study of perception; indeed, as Geertz suggests, the possibility remains
that the most crucial questions about categorization may continue to elude
cognitive-scientific methods. Perhaps cognitive science as a field will prove
unequal to the task of explicating the genesis, nature, and use of our more
complex and evocative concepts and categories.
Yet, even if issues of categorization turn out to be susceptible to a
cognitive-scientific analysis; even if (say) the naturalistic view turns out
to be fully vindicated, there will remain crucial unresolved questions. For
classification or categorization is ultimately a tool-a means whereby in-
dividuals organize their world so as to solve certain problems and to
achieve certain ends. The ultimate achievements of this field will depend
on whether the principles evolved to deal with relatively simple concepts
and modest classification will prove applicable to complex concepts and
systems of concepts.
Just what sort of an organism will utilize these capacities, and toward
what ends will they be used? Here cognitive scientists have found them-
selves confronting value-laden aspects of human cognition-and, in par-
ticular, have directed considerable attention to the question whether in-
dividuals can be thought of as proceeding in a logical or rational fashion.
This question is of importance for scientific reasons, since much of cogni-
tive science is based on the model of the logical computer; it also has
important political and social implications. The answers to questions of
human rationality, with which I shall deal in the following chapter, turn
out to be surprising, and not particularly reassuring to those who place
their faith in reason.
359
13
How Rational a Being?
Whenever there is a simple error that most laymen fall for,
there is always a slightly more sophisticated version of the
same problem that experts fall for.
-AMos TVERSKY
While the Greeks sometimes defined man as a featherless biped, they had
far more invested in another definition: man as rational being. Only hu-
mans can invoke a form of thought that adheres explicitly to rules and
leads to conclusions that can be judged by the community as valid or
invalid. Unpacking this Greek notion a bit, it proves possible to distinguish
between the individual who reaches valid conclusions-and can therefore
be deemed rational-and the procedure by which these conclusions are
reached-a procedure termed logic. And so, one might conceive of a crea-
ture who is rational but reaches conclusions by means other than logic-
for example, by shrewd intuition, lucky guessing, or being programmed to
issue only valid responses.
Though surrounded (in a sense) by evidence of the irrationality of
human beings, philosophers have clung to the notion that human beings
are logical and rational-or at least to the ideal that human beings should
strive for rationality and that they have the potential to achieve it. This
preoccupation with an ideal of rationality is hardly surprising, since phi-
losophy itself has sought to proceed by rational means. Indeed, Bertrand
Russell once speculated that Aristotle was the first to define man in terms
of rationality (quoted in Pylyshyn 1984, p. 257). Over the years, the
development of the field of logic has been closely intertwined with the
history of philosophy: just as the logic of Aristotle's time informed Greek
philosophy so the logics of Frege, Whitehead and Russell, and Kripke have
informed the philosophy of today.
360
Cognitive science was conceived in the shadow of contemporary logic.
As I noted in discussing the Hixon Symposium in chapter 2, both early
work on computers and the model of the neuron as a logical circuit encour-
aged a view of thought as logical (Jeffress 1951). Furthermore, the first
generation of cognitive scientists embraced a model of human beings that
was decidedly rationalistic. Jerome Bruner, Herbert Simon, Allen Newell,
and Jean Piaget all elected to investigate issues involving the abilities of
human beings to reason validly. In fact, the problems solved by the first
computer programs were problems in logic; the classification tasks investi-
gated by psychologists required logical deductive processes; Piaget went
one step further, not only studying problems of logic, but assuming that
"developed" humans reason by invoking principles of logic. As he once
put it, "reasoning is nothing more than the propositional calculus itself"
(Inhelder and Piaget 1958, p. 305).
But the faith of the initial generation in a study of logical problems
and its determined search for rational thought processes may have been
misguided. Empirical work on reasoning over the past thirty years has
severely challenged the notion that human beings-even sophisticated
ones-proceed in a rational manner, let alone that they invoke some logical
calculus in their reasoning. Once again, a computational age has docu-
mented departures from computerlike precision. This realization has
seeped through many lines of research, but few would quarrel with my
selection of Philip Johnson-Laird (from Great Britain) and the team of
Daniel Kahneman and Amos Tversky (originally from Israel) as the
spokesmen for this particular chapter of cognitive science. Not only have
these scholars developed stunning demonstrations of human departures
from rationality; they have also offered explanations of the reasons we
humans often go wrong in the ways we do.
The Illogic of Human Reasoning
Cards with Numbers

Consider the following problem developed by Peter Wason and Philip
Johnson-Laird (Johnson-Laird and Wason 1970; Wason 1966; Wason and
Johnson-Laird 1972). Four cards are laid out with their faces displaying,
respectively, an E, a K, a 4, and a 7. You are told that each card has a letter
on one side and a number on the other. You are then given a rule, whose
truth you are expected to evaluate, "If a card has a vowel on one side, then
361
I I I I TowARD AN INTEGRATED CoGNITivE SciENCE
it has an even number on the other." You are then allowed to turn over
two, but only two, cards in order to determine whether the rule is correct
as stated.
If you have already met this problem, or one of its close relatives, you
should have little trouble getting it correct. (In this case, your memory is
being tested, not your powers of reasoning.) But if this is your maiden
encounter, you will almost certainly miss it, as have over 90 percent of
subjects (including many logicians) to whom it has been presented in many
settings. Most subjects realize that there is no need to select the card
bearing the consonant, since it is clearly irrelevant to the rule; they also
appreciate that it is essential to turn over the card with the vowel, for an
odd number opposite would infirm the rule. The difficulty inheres in
deciding which of the two numbered cards to pick up. There is a strong
temptation to pick up the card with the even number, because the even
number is mentioned in the rule; and this temptation proves fatal to a
majority of subjects. But, in fact, it is irrelevant whether there is a vowel
or a consonant on the other side, since the rule does not actually take a
stand on what must be opposite to even numbers. On the other hand, it
is essential to pick up the card with the odd number on it. If that card has
a consonant on it, the result is irrelevant. If, however, the card has a vowel
on it, the rule in question has been infirmed, for the card must (according
to the rule) have an even (and not an odd number) on it.
The fact that this problem proves hard (even though, once explained, it
seems evident enough) gives one pause: ready explanations for this result
elude those who place great faith in the logical capabilities of their fellow
humans and themselves. But to my mind there is an even more interesting
twist to theW a son and Johnson-Laird demonstration. Consider the follow-
ing problem, again using four cards, each representing a journey. Each card
has a destination on one side and a mode of transportation on the other. This
time the cards have printed on them the legends, respectively, "Manches-
ter," "Sheffield," "Train," and "Car"; and the rule is: "Every time I go to
Manchester, I travel by train." While this rule is formally identical to the
number-letter version, it poses relatively little difficulty for individuals. In
fact 80 percent of subjects realize the need to turn over the card with the
word "car" on it. Apparently, one realizes that if the card with "car" on it
has the name "Manchester" on the back, the rule is infirmed; whereas it is
immaterial what it says on the back of the card with" train" on it since, as far
as the rule is concerned, one can go to Sheffield any way one wants.
Why is it that 80 percent of subjects get this problem correct, whereas
only one tenth know which cards to turn over in the logically identical
vowel-number version? Possibly a person is much better at solving a
problem entailing familiar material-material that enables one to place
oneself inside the situation, to figure out what one would do, what it
362
would be reasonable to do. Indeed, you are more likely to succeed on the
aforementioned problem if you are British (and hence know of Sheffield
and Manchester) and if you are of an older generation (and so have grown
up with trains as a likely mode of travel).
The results clearly challenge any notion that individuals are "logic
machines," capable of applying the same modes of reasoning, independent
of the specific information in a problem. Thus, humans are notably different
from an ideal computer: the actual contents of a problem (vowels versus
vehicles) cannot make any difference for such a syntactically governed
apparatus. Yet just as the particular structure of each category colors a
subject's classifying strategies in a Roschian experiment, so, too, the content
of specific problems determines how all of us (except, perhaps, highly
trained logicians) will proceed on an apparently simple reasoning task.
Results on these and many other puzzles have convinced Philip John-
son-Laird, a leading cognitive scientist, that people do not employ a mental
logic in solving problems (1983). The kinds of logic described by logicians
simply seem irrelevant to normal individuals. We do not construct truth
tables and look up the result: we do not use formal rules of inference. But
Johnson-Laird maintains that there can be reasoning without logic. The
dilemma facing the cognitivist is to allow both for rationality and for
human error.
Puzzles about traveling or games with vowels and numbers may seem
a shade artificial, but there is nothing contrived about the syllogism. This
form of reasoning goes back at least to Greek times and was considered by
Aristotle (often our judge in these matters) as the core of logic. It is used
unreflectively by ordinary individuals as part of daily experience; and as
Johnson-Laird neatly shows, it is even drawn upon by critics who wish to
belittle the importance of syllogisms. One might say, "Syllogisms are
artificial." Furthermore, "Psychologists shouldn't study things that are
artificial." Therefore, "Psychologists should study the kinds of inferences
that are used regularly in daily life." In the very act of dismissal, this critic
has invoked syllogistic reasoning.
Artists and Beekeepers

According to Johnson-Laird's analysis, there are sixty-four kinds of
syllogism-all variations of the following example:
All the artists are beekeepers. (All A's are B.)

All the beekeepers are chemists. (All B's are C.)
The subject (and anyone can play) is asked to figure out which conclu-
sions, if any, follow from these two premises. Johnson-Laird and his col-
363
leagues have found that such syllogisms pose considerable difficulties; and
that, often as not, people will draw invalid conclusions from them. More-
over, explaining the patterns of success and failure on such problems has
proved difficult for the logicians and psychologists who have sought a
comprehensive account. These misadventures have resulted because ex-
perts on syllogistic reasoning assume that individuals untrained in formal
logic nonetheless use some variant of it (for example, visual aids like Euler
circles or Venn diagrams) to solve the problems.
These experts have evolved elaborate accounts purporting to show
how the average person approaches syllogisms. These methods can per-
haps work in principle-if, for example, one has unlimited memory capac-
ity and has mastered sophisticated mathematical notations; but Johnson-
Laird illustrates convincingly that even the talented college student cannot
proceed by embracing these methods: "The present theories are too fragile
to bear the weight of human reason" (1983, p. 93). Either the theorists
succeed in accounting for deductive errors but fail to account for the
rationality that is exhibited by subjects, or they explain the ability of
subjects to reason adequately under ideal conditions but fail to illuminate
the kinds of error made by human subjects. The trick is to offer an explana-
tion that can account for both human rationality and human error. John-
son-Laird therefore introduces his major contribution to cognitive science:
the notion of a mental model.
He asks us to pretend that we have the power to conjure up individu-
als who fulfill one or more of the roles stated in the premises. Clearly, we
can create individuals who are at one time artists, beekeepers, and chemists
-who exhibit the trio of roles diSZussed in the premises. These individuals
can be represented in mental model form as
artist-beekeeper-chemist
We also know that there are individuals who are beekeepers and chemists
but not artists:
beekeeper-chemist
beekeeper-chemist
beekeeper-chemist
And we know from common sense (as well as from careful consideration
of the problem) that there can be chemists who are not beekeepers:
chemists
chemists
chemists
364
We now have created a mental model that arrays all of the informa-
tion that can be taken directly from the premises given. We see that there
are three categories of individual: those who represent all three occupa-
tions; those who are beekeepers and chemists; and those who are just
chemists. (The fact that I have listed three instances in each category is
accidental; one may conjure up as many exemplars as one wishes.) Reading
off of these arrays, we can now proceed to answer questions or draw
conclusions. For example, if we want to determine whether all the artists
are also chemists, we can simply look at the tableaus and confirm that all
the artists are indeed chemists.
Johnson-Laird's approach is important in that it proceeds differently
from other tacks, which in some way follow the laws of formal logic, either
explicitly or implicitly. There is no need to translate the premises, implic-
itly or explicitly, into the p's and q 's of formal logic, to say that all p's are
q's, and all q's are rand to deduce, according to an algorithm, what follows
from that array. Nor is there any need to draw overlapping circles (the
visual embodiment of formal logic in a system like Euler circles) and to
read off which aspects overlap one another under each circumstance.
Rather, the individual engaged in constructing a mental model is simply
employing whatever medium is comfortable to him (words, images, some
hybrid) in order to represent the information for himself in a convenient
and readily accessible manner.
According to Johnson-Laird, one first represents the individual tok-
ens (artists, beekeepers) in some manner and thereby forms a mental
model of the first premise. Next, one adds the information in the second
premise (that beekeepers are chemists) to the mental model of the first
premise, taking into account the different ways in which this can be
done. An advisable strategy here is to establish the possible arrange-
ments (whether an individual can be a beekeeper without being an artist)
using as few "imaginary actors" as possible, though, as I have noted, the
precise number of actors depicted proves irrelevant to the syllogistic in-
ferences that are eventually drawn. Finally, after all of the mental pic-
tures have been constructed, an integrated set of pictures is submitted to
a test: a search is undertaken for an interpretation of the premises that is
inconsistent with the model. An inference is valid if, and only if, there is
no way of interpreting the premises that is consistent with a denial of
the conclusion.
In the example just given, it is possible to arrive at a valid conclusion
simply by constructing one mental model. Thus the solution of this prob-
lem is relatively simple: one avoids the explosion of combinatorial pos-
sibilities that plague conventional formal approaches to syllogistic reason-
ing. Of course, some other problems prove far more complex to model. For
example, Johnson-Laird introduced the following syllogistic premises:
365
All the bankers are athletes.

None of the councillors are bankers.
If you believe that this syllogism is easy, you are almost certain to be
wrong. (Even if you know it is hard, you will probably fail.) In fact, as
Johnson-Laird shows, there is only one fool-proof conclusion:
Some of the athletes are not councillors.
This conclusion is difficult to draw correctly because it requires a subject

to devise three separate mental models and then to integrate them so as
to reach a single conclusion. Of the sixty-four possible syllogisms investi-
gated by Johnson-Laird, four are as fiendishly difficult as this one: they
require individuals to construct and evaluate three separate models and, to
make matters worse, fail to exhibit certain surface or "figural" cues that
ordinarily help one to solve syllogisms.
So far what Johnson-Laird says seems reasonable, but why accept it
as the best account of the thinking involved in syllogistic reasoning? He
sets out several criteria for a theory of reasoning and proceeds to show how
his approach meets these criteria. First of all, it is necessary to have a way
of characterizing all sixty-four kinds of syllogism and to predict which will
pose the most problems for subjects. Empirical work with subjects has
confirmed the basic outlines of this claim. Second of all, it is necessary to
lay out the requisite processes by which individuals solve these problems
and then differentiate among the most and the least successful subjects in
terms of whether they can exhibit each of the hypothesized component
skills. Johnson-Laird's model predicts that the decline in performance re-
flects the number of separate models that have to be constructed. A third
desideratum is the capacity of the theory to explain how children handle
syllogisms and to pinpoint the factors that will improve their performance.
It turns out that young children cannot handle syllogisms that require
more than one model; indeed, children succeeded best in that form of
syllogism which Aristotle considered perfect:
All the A are B.

All the B are C.
Therefore all the A are C.
All of these research maneuvers are carried out from the perspective
of psychology, though admittedly a psychology informed by a sophis-
ticated grasp of the ways and traps of logic. But Johnson-Laird penetrates
into the heartland of cognitive science because (like Stephen Kosslyn
working in the area of imagery) he has implemented his theory via a
computer program. The computer program consists of three basic steps:
366
1. Construct a mental model of the first premise.
2. Add the information in the second premise to the mental model of the first
premise, taking into account the different ways in which this can be done. (This
turns out to be the trickiest aspect of the process, the one that determines whether
one, two, or three different models need be constructed.)
3. Frame a conclusion to express the relation, if any, between the "end"
terms that hold in all the models of the premises.
These steps constitute what Johnson-Laird terms an effective proce-

dure-a procedure that, if carried out rigorously, guarantees that one will
reach the appropriate conclusion. In his view, cognitive science rests upon
discovery of such effective procedures which can be carried out not only
by those imperfect mechanisms called human beings but also by those
impeccable mechanisms called digital computers. Johnson-Laird's proce-
dure worked perfectly on his computer and thus is effective.
Johnson-Laird attributes the power of his theory to the fact that it is
compatible both with the sorts of errors that subjects make and with a
completely valid set of deductions. He concludes:
There appears to be no branch of deductive inference that requires us to

assume the existence of a mental logic in order to do justice to the psychological
phenomena. To be logical, an individual requires, not formal rules of inference, but
a tacit knowledge of the fundamental semantic principle governing any inference:
a deduction is valid provided that there is no way of interpreting the premises
correctly that is inconsistent with the conclusion. Logic provides a systematic
method for searching for such counter-examples. The empirical evidence suggests
that ordinary individuals possess no such methods. (Quoted in Mehler, Walker,
and Garrett 1982, p. 130}
Mental Models as a Panacea?

Syllogisms are but one of several areas examined by Johnson-Laird in
his effort to demonstrate the power of his mental-models approach. In his
recent book Mental Models (1983), he applies his approach with impressive
success to such topics as inference, word meaning, grammar, and the com-
prehension of discourse. In each case, he describes the nature of the prob-
lem to be solved, the competing explanatory models (including models
based on propositions, schemas, images, and other staples of a cognitive-
scientific diet), and attempts to show why the mental-models approach
illuminates the processes exhibited by human beings and can be effectively
realized on a computer. This is more than a series of effective demonstra-
tions: it is a persuasive argument for a certain approach to cognitive sci-
ence. Mental Models may well serve as a mental model for the next genera-
tion of cognitive scientists.
When it comes to describing what a mental model is and how it differs
367
from other competing accounts, Johnson-Laird is less than completely

successful. It may be that he is simply placing too great a demand on a
single concept-stating, for example, that "all our knowledge of the world
depends on mental models" (1983, p. 419). As one sympathetic critic has
declared, "Johnson-Laird succumbs to the temptation to see his nascent
theory of mental models as the solution to everyone else's problem" (Stich
1984, p. 189).
Johnson-Laird seems to sense this difficulty, though he may view it
as less of an obstacle than an opportunity. He embraces the versatility of
such models:
Since mental models can take many forms and serve many purposes, their
contents are very varied. They can contain nothing but tokens that represent
individuals and identities between them, as in the sorts of models that are required
for syllogistic reasoning. They can represent spatial relations between entities, and
the temporal or causal relations between events. A rich imaginary model of the
world can be used to compute the projective relations required for an image.
Models have a content and form that fits them to their purpose, whether it be to
explain, to predict, or to control. (1983, p. 410)
Despite these probably excessive and perhaps unsustainable claims,

mental models tum out to be useful devices for accounting for a subject's
behavior in delimited domains, like syllogistic reasoning. First of all,
Johnson-Laird's approach explains how individuals can succeed on some
questions-and also can fail on others-without the need for invoking
formal logical methods. Thus, it can account for a wide range of perfor-
mances. Second of all, the various predictions built into the model have
been tested not only with human subjects but also with a computer pro-
gram. Again, while a computer program is no guarantor of success, the
fact that simulations have proved workable indicates that at least there
are no hidden errors or contradictions in the theory. In its own terms, the
model works.
In his discussion of rival explanatory modes in the cognitive sciences,
Johnson-Laird makes a telling point. In his opinion, it is too simple to say
that images are (or are not) modes of representation, to argue (as have
scholars like Pylyshyn [1984)) that all representation is best thought of in
propositional forms, or to maintain (as has John Anderson [1978)) that it
is impossible in principle to decide whether a process entails propositions
or images.* Johnson-Laird prefers to posit at least three types of mental
representation: propositional representations that resemble natural lan-
• Anderson has constructed a proof showing that it is always possible to construct an
alternative account using a different sort of representation that behaves in an equivalent way
(1978).
368
How Rational a Being1
guages; mental models which are structural analogues of the world; and
images that are the perceptual correlates of models from a particular point
of view (1983, p. 165). We ought to think of individuals as representing
information at several different levels of abstraction: moreover, the form
of representation at one level need not be the same as the form of represen-
tation at another level. Just as a computer may have several different
languages, ranging from machine code to a high-level programming lan-
guage, so, too, a psychological process might use only strings of symbols
at one level but involve images or mental models at a higher level of
representation. This ecumenical approach seems to do more justice to the
wide range of human psychological processes brought to bear on such
problems as text understanding or inference than does a stubborn adher-
ence to a single mode of mental representation.
Returning to the puzzle of logical reasoning, we find that Johnson-
Laird has enriched our understanding in two ways. To begin with, he has
shown that the ways in which human beings were once thought to ap-
proach problems of reasoning simply do not hold: one does not reason as
classical logic would suggest. Yet humans are not irrational either. Through
a careful analysis of syllogistic reasoning, the conduct of experiments with
subjects of different ages and degrees of expertise, and the simulation of
his model on a computer, Johnson-Laird has arrived at a picture of reason-
ing that seems robust and viable. A combination of several of the cognitive
sciences has yielded a genuine clarification of a long-standing philosophi-
cal issue-the degree of rationality of human beings.
Why does logic fail? As anthropologist Roy D' Andrade has noted, the
vocabulary of the logician is a second-order vocabulary (1982). It is a
statement not about things or events but, rather, about the consistency or
the inconsistency of statements. Ordinary language statements do not
usually refer to truth conditions but refer rather to states of affairs in the
world: people appear to be so designed (or so educated) that their major
interest focuses on what can happen in the world under such-and-such
conditions. The soundness, the speed, and the complexity of the reasoning
that individuals exhibit seem primarily a function of the degree of famil-
iarity and organization of the materials being processed, rather than a
function of any special or general ability of the person doing the reasoning.
And so there are appreciable differences in how a given person can reason
about different topics, topics that, from a formal point of view, call upon
the same degree (and even principles) of logical expertise.
This line of reasoning drawn from the analyses of Johnson-Laird,
Wason, and D'Andrade suggests that we can better understand the logical
reasoning of humans not by imputing to them any formal logical calculus
but by attending instead to two factors. The first has to do with content:
369
the greater the familiarity and the richer the relevant schemata which are
available, the more readily can one solve a problem. The second attribute
has to do with form: one succeeds on problems to the extent that one can
construct mental models that represent the relevant information in an
appropriate fashion and use these mental models flexibly. Just how one
learns to construct such mental models, to integrate them with "real
world" knowledge, and to deploy them appropriately in the proper
circumstances are fertile questions for developmental and educational
psychology.
If Johnson-Laird and company are correct, the kinds of principles
devised by logicians-and invoked by researchers like Piaget-will turn
out to have only limited applicability to how we reason in the real world.
Apparently we have evolved as creatures who are most likely to succeed
on tasks that contain familiar elements and allow the ready construction
and manipulation of mental models. Considerations of pure logic, a field
that developed long after our survival mechanisms had fallen into place,
may be useful for certain kinds of information under certain circumstances
by certain individuals. But logic cannot serve as a valid model of how most
individuals solve most problems most of the time.
Biases in Human Cognition:

The Tversky-Kahneman Position
As if the onslaught from Johnson-Laird and colleagues were not enough,

still other attacks on the rationality of human beings have been launched
in recent years by imaginative (if not insidious) researchers. A well-known
set of studies was carried out in the 1970s by Amos Tversky and Daniel
Kahneman, then of the Hebrew University in Jerusalem. Once again, the
upshot of these studies was that individuals often performed radically
differently on problems that had the same formal structure; moreover,
these performances reflected strong biases in the human cognitive system
which must be taken into account in any theory of human reasoning. Thus,
while Johnson-Laird has illuminated the kinds of processes used by in-
dividuals in solving classical logic problems, Tversky and Kahneman have
singled out the kinds of heuristics and strategies that guide-if they do not
determine-the ways in which individuals reason in everyday life (Kahne-
man, Slovic, and Tversky 1982; Kahneman and Tversky 1982, 1984;
Tversky and Kahneman 1983).
370
Theater Tickets and Coin Tossers
Let me consider a few of the numerous intriguing examples reported

by Tversky and Kahneman: Asked to say which sequence of heads and
tails is less common, HHHTTT or HTHHTT, most individuals will choose
the former. Apparently individuals judge frequency of samples on the
basis of their similarity to the features of the "parent" population. Since
irregularity is an essential feature of randomness, irregular samples are
judged more likely than regular samples, even though both series are
equally likely to occur.
Imagine that you are on the way to a Broadway play with a pair of
tickets which cost forty dollars (this is some years ago!) and discover you
have lost the tickets. Would you pay another forty dollars? Now imagine
that you are on the way to the theater to buy these tickets. Upon arrival,
you realize that you have lost forty dollars in cash. Would you now buy
tickets to the play? Clearly, on an objective basis, the two situations are
identical because in both you are forty dollars in the hole. Nevertheless,
most people report that they would be more likely to buy new tickets if
they had lost the money than if they had lost the tickets. The argument
is that the same loss is assigned to different "mental accounts": the loss of
forty dollars in cash is entered into an account distinct from the play and
so has comparatively little effect on whether one buys new tickets. On the
other hand, the cost of the lost tickets is attributed to the account of
"attending the theater": one is loath to accept the doubling of the cost of
the play-shelling out eighty dollars for a pair of tickets.
Another example: Mr. Crane and Mr. Thomas are scheduled to leave
the airport at the same time, though on different flights. They travel to the
airport in the same limousine, are caught in a traffic jam, and arrive at the
airport a half-hour after their scheduled departures. Mr. Crane is told that
his flight left on time, while Mr. Thomas is told that his flight was delayed
and left just five minutes ago. Nearly everyone agrees that Mr. Thomas
will be more upset, even though in fact the two men's objective conditions
are indistinguishable-after all, both missed a plane. The reason, in the
view of Tversky and Kahneman, is that in the play of one's imagination,
Mr. Thomas comes much closer than Mr. Crane to catching his flight, and
thus the frustration experienced is greater. Similarly, an individual whose
lottery ticket differs by only one cipher from that of the winning ticket is
much more upset than the individual whose number is remote from the
winning entry.
Imagine that you are about to buy a jacket for $125 and a calculator
for $15. The calculator salesman tells you that the calculator you want to
buy is on sale at the other branch of the store, twenty minutes away, for
371
$10. Would you make the trip? Most people say that they will. Another
group is asked a similar question. This time the cost of the jacket is changed
to $15, and the cost of the calculator to $125 in the original store and to
$120 in the branch. Of respondents presented with this version, the major-
ity said that they would not make the extra trip. Note that in both cases
the total purchases are the same: the choice is always whether to drive
twenty minutes to save $5. But apparently respondents evaluate the saving
of $5 in relation to the price of the calculators. In relative terms, a reduction
from $15 to $10 (or 50 percent) is less resistible than a reduction from $125
to $120 (less than 5 percent).
Finally, meet Linda. She is thirty-one years old, single, outspoken, and
very bright. She majored in philosophy. As a student, she was deeply
concerned with issues of discrimination and social justice and also par-
ticipated in antinuclear demonstrations. You are now asked to rank, from
most to least probable, a series of eight statements. Included in the list are
the statements: "Linda is a psychiatric social worker," "Linda is a bank
teller" and "Linda is a bank teller and is active in the feminist movement."
On any rational account, it is more probable that Linda is a bank teller than
that Linda is both a bank teller and active in the feminist movement. The
probability of x, after all, is always greater than the probability of indepen-
dent event x and independent event y. Yet more than 80 percent of sub-
jects, including those who are sophisticated in statistics, assent more read-
ily to the statement that Linda is a bank teller and a feminist than to the
statement that Linda is a bank teller.
Why this clear flouting of rationality? Tversky and Kahneman see the
laws of probability being overwhelmed by the principle of representa-
tiveness. People are keenly attuned to the likelihood that someone with
certain characteristics will also exhibit other ones. (To the extent that
someone is a social activist, she is likely to be a feminist.) Indeed, the more
supporting details are consistent with this representative portrait, the more
likely a subject will assent to these new details, even though (on a rational
basis) the added details increasingly constrain the set and hence render it
less probable in absolute terms. Thus, given the information that Linda is
a certain kind of person, subjects readily fit in other events that have in
the past been representative of such persons, and in the process ignore
what they otherwise know about probability.
And this "otherwise" knowledge does still exist. Asked the abstract
question "Which is more probable, x alone or x and y ," subjects readily
consent that x alone is more probable. Moreover, when confronted with
the apparent contradiction between this abstract response and the Linda
question, they readily admit that they have made an error. Thus, it is too
simple to say that subjects were simply fooled: the response based on
372
representativeness in the Linda problem seems to reflect a deep-seated bias
in human judgment.
These are but a few of the many vivid instances studied by Tversky
and Kahneman and, of the lot, among the easiest to present.* But these
examples should underscore the point that individuals do not reach deci-
sions in a way that is logically consistent or that obeys the laws of proba-
bility. I do not mean, however, that the behavior is illogical or inexplicable.
Rather, Tversky and Kahneman have revealed a separate psychology of
preferences which does not follow strictly from an economic calculation
of gains and losses but instead focuses on how individuals "frame" selec-
tions. Among the relevant considerations are: whether a person construes
the situation as one in which there is a guaranteed loss (as opposed to a
likely loss); whether a result will have a substantial effect on one's custom-
ary style of life; whether one has ready access to an instance of the category
in question; how one's imagination plays on what might have happened;
how closely an example resembles a prototype about which one already
holds strong views; the mood that one expects to be in; and the way in
which a question is actually phrased (for example, does one speak of the
killing of four hundred out of six hundred persons, or of the sparing of two
hundred out of six hundred?).
Overall, Tversky and Kahneman conclude that statistical principles
and rules of deduction are simply not imported from the kit of the mea-
surement scientist into the reasoning of everyday life. People are not
coherent in the way that it would be nice to believe they were. In even-
tempered tones, these investigators conclude, "The descriptive study of
preferences also presents challenges to the theory of rational choice be-
cause it is often far from clear whether the effects of decision weights,
reference points, framing, and regret should be treated as errors or biases
or whether they should be accepted as valid elements of human experi-
ence" (Kahneman and Tversky 1982, p. 171).
A Philosophical Critique
The results reported by Tversky and Kahneman, as well as those
gathered by Johnson-Laird, Wason, and other workers, have not gone
unnoticed ~r unre~arked upon by those keepers of human rationality-
the professzonal philosophers. One such is L. Jonathan Cohen of Oxford
University, who has taken on Tversky, Kahneman, and their colleagues in
a series of critiques. Cohen (1981) takes three swipes at the notion that
most individuals do not behave in a logical or rational fashion. The first
. •Interested re~ders can play subject by consulting the dozens of examples recounted
m Kahneman, Slov1c, and Tversky 1982.
373
III I TowARD AN INTEGRATED CoGNITIVE SciENCE
line of argum~nt simply holds that what is rational should be determined

by ~h~ .man m the street; he is the ultimate arbiter, and findings that
statis~ICla~s or experimenters may choose to impose on laymen are, strictly
speakmg, Irrelevant. On this line of argument, individuals can make errors,
of course; but except for such problems of "performance," it ought to be
assumed that the average person is competent. After all, no one questions
judgments about grammar as rendered by the man on the street and
judgments of reason should be similarly honored. This line of arsu'ment
may have a certain persuasiveness; but, in the last analysis, it strikes me
as antiscientific since it essentially closes the discussion. After all, if every
human (of whatever age or mental state) is considered rational, no matter
how he or she happens to reason, and no matter whether he or she concurs
with others, then there is no way to study rationality at all. It is ubiquitous
and unchallengeable.
A second line of argument maintains that it is unreasonable to expect
ordinary individuals to be aware of laws of probability and other statistical
regularities: most of the claimed findings are seen as reflecting a lack of
education or of special knowledge, rather than deficiencies of cognition.
On this line of reasoning, we would all be superrational if we were to
receive expert training. No one has yet offered to train John Q. Public on
the laws of probability, but there is already plenty of anecdotal evidence
and some experimental evidence to suggest that even trained experts ex-
hibit the kinds of bias uncovered by Tversky and Kahneman (Kahneman,
Slovic, and Tversky 1982).
The third-and, to my mind, the most interesting-line of argument
actually challenges the relevance altogether of laws of statistics like Bayes'
theorem. According to the theorem, postulated in 1763 by Thomas Bayes,
one should take into account baseline (or base) rates when making judgments
about probability in a specific situation. Thus, if one is trying to decide
whether a given verbal portrait pertains to a lawyer or to a trial lawyer,
it is pertinent to take into account the fact that there are far more lawyers
in a society than there are trial lawyers.
Cohen directs his critical sights on the following example from the
published research of Tversky and Kahneman. Subjects were told that, in
a certain town, blue and green cabs operate in a ratio of 85 to 15, respec-
tively. (In Bayesian lingo this is the so-called base rate.) A witness has
identified a cab in a crash as green; and the court is told that, under the
relevant light conditions, such a witness can reliably distinguish blue ca~s
from green cabs in 80 percent of cases. Subjects were then_ asked: ~hat 1s
the probability (expressed in percentages) that the cab mvolved m the
accident was actually blue?
The median response in this task was 0.2. Investigators like Kahneman
74
How RaHonal a Being?
and Tversky claim that this response reflects a serious error because it
indicates a failure to take into account the base rates-that is, the prior
probabilities. The investigators reason as follows. The fact that far more
cabs are blue than green must be taken into account in computing the
likelihood that the cab was, in fact, green. They then make four computa-
tions: the likelihood of correct identifications of the cab as blue, given that
it is blue (85% X .8 = 68% ); the incorrect identification of a cab as blue,
given that it is actually green (15% X .2 = 3% ); the correct identification
of a cab as green, given that it is actually green (15% X .8 = 12%); and
the incorrect identification of the cab as green, given that it is actually blue
(85% X .2 = 17% ). All together, the identifications of the cab as green
will be 29 percent (17% + 12%) and the fraction that is wrong will be
17/29. Consequently, according to the experimenters' mode of reasoning,
the probability that the cab in the accident was actually blue is 17/29 and
not 1/5.
But, argues Cohen, one need not approach this problem in this text-
book way (in this case via Bayes' theorem). He claims, for instance, that
jurors ought not to rely on probability if they can avoid doing so: such an
invoking of probability assumes that the issue before the court concerns
a long run of cab-color-identification problems, where actually the juror
is only trying to decide about this particular case. Strictly speaking, jurors
are here occupied with the probability that the cab actually involved in the
accident was blue, on the condition that the witness said it was green. Since
the jurors know that only 20 percent of the witnesses' statements about
cab colors are false, they rightly estimate the probability at issue as 1/5,
without violating Bayes' law. The fact that cab colors actually vary by an
85-to-15 ratio may properly be considered irrelevant to the estimate: after
all, this fact neither raises nor lowers the probability of a specific cab-color
identification being correct on the condition that it is an identification by
this particular witness. As Cohen phrases it, "A probability that holds
uniformly for each of a class of events because it is based on causal
properties, such as the physiology of vision, cannot be altered by facts,
such as chance distributions that have no causal efficacy in the individual
events" (1981, pp. 328-29).
Cohen comes up with another example that strikes much closer to
home. Suppose you are suffering from a disease which is either A or
B. It is known that A is nineteen times as common as B. The two diseases
are equally fatal if untreated, but it is dangerous to combine treatments.
Your physician orders a test which turns out to be correct on 80 percent
of the cases where a differential diagnosis must be made. The test reports
that you are suffering from the much rarer disease, B. Should you nonethe-
less opt for the treatment appropriate to A? According to the Bayesian
375
I I I 1 TowARD AN INTEGRATED CoGNITIVE SciENCE
view, this would be the proper step. In view of the prior distribution of
the disease, the probability that you are suffering from A will be 19/23.
But you might choose to ignore the prior probabilities, on the supposition
that the probability of your suffering from B is 4/5. According to Cohen,
it is irrational for you as a patient to answer in terms of the overall
probabilities rather than in terms of the particular diagnosis just made.
Paradoxically, if you followed the point of view in the statistical textbook,
it would be a waste of time even to apply the test for the disease, since
whatever its results, it is still more prudent on a statistical basis to assume
that you have disease A, the statistically more common one.
Cohen discerns different interests at stake here. The administrator
who wants to secure a high rate of diagnostic success for a particular
hospital at minimal cost would be well advised to follow the laws of
general probabilities, and so should eliminate the test. But the patient is
concerned with success in his own particular case, not with the probabilis-
tic success of the system in the long run, and so needs to evaluate test
results with respect to his chances. Long-run strings of success are of no
relevance to him.
In suggesting that the factors affecting the promulgators of a statistical
law are not identical to those influencing a solitary individual making a
decision about his or her own life, Cohen makes an interesting move. But
it has not won him much favor with other experts on judgment, such as
those asked to comment on his article in The Behavioral and Brain Sciences
(Cohen 1981). Cohen was castigated as an apologist for errors and for the
status quo, as a person who would undo the learning of science, and who
himself was incapable of putting forth an argument or reasoning coher-
ently and thus was dependent upon unsupported assertion and illusion.
These criticisms struck me as being excessively severe, as though Cohen
had hit a vulnerable spot. But to my mind, rather than demonstrating that
people are logical (as he has intended to do), Cohen has, in fact, provided
further evidence that they depart from logic and rationality in certain
systematic ways. He also has reminded us that acting in one's own ad-
judged self-interest is not necessarily the same as making a judgment
about which outcome is more likely to happen in some hypothetical event.
Some further points can be made as well. First of all, individuals have
evolved in order to survive in a certain range of biological and social
environments, not to satisfy some abstract notion of logic. It may well be
that the kinds of induction and deduction individuals habitually (and even
reflexively) make are most likely to lead to survival. And so they are
rational in this absolute sense, even if they appear irrational in the light
of certain textbook principles.
Another line of argument underscores the kinds of bias that have been
376
built into favored psychological paradigms. To the extent that experiments
harbor illusions or tricks they may be seen as having limited relevance to
most daily behavior. Certainly if one slips traps into questions, it is easy
enough to secure wrong answers or answers that make subjects look fool-
ish. Piaget and his colleague Barbel Inhelder once designed a test in which
one gives young children a set of five tulips and two roses and then asks,
"Are there more tulips here or more flowers?" (Inhelder and Piaget 1964).
Until the age of seven, children will answer that there are more tulips,
seemingly oblivious to the fact that the term flowers subsumes all of the
tulips and all of the roses. Piaget and Inhelder confidently concluded that
young children are unable to compare a whole set (flowers) with one of
its subsets (the tulips). But as philosopher Jonathan Adler has pointed out
(1984), it may be too simple merely to assume that children are incapable
of making such comparisons. Rather, the children may simply not be
expecting the kind of question that violates ordinary conversational rules
of being "relevant" and pertinent. Children may just assume that the
comparison being asked for is the perceptually evident one-Are there
more tulips or more roses?-rather than the obscure if not downright
deceptive query, Are there more tulips or more tulips plus roses? On this
analysis, if children are asked for comparisons where a subclass could be
more legitimately compared with the entire class, they would answer cor-
rectly. And, indeed, it turns out that when such questions are posed to
children-Are there more children or more people?-they emerge as sensi-
tive to the principles of "class inclusion" (see also Donaldson 1978).
Applying this perspective to the preceding examples, we can see that
individuals might well be apprehending the Tversky-Kahneman problems
in light of the ways in which people usually converse and the kinds of
information they usually exchange. In the "Linda-bankteller" problem, for
example, one assumes that the information about Linda's previous political
involvements would only have been given if the listener was supposed to
draw the conclusion that she was a feminist. Thus, a listener simply makes
the reasonable assumption that the speaker wants him to believe that
Linda is a feminist--else why trouble to mention so much about her
political and social attitudes? In the problem about the taxicab, on the
other hand, it might be assumed that the information about base rates has
no relation to the question at hand. For as Tversky and Kahneman have
themselves pointed out, subjects respond differently when told, "Al-
though the two companies are roughly equal in size, 85 percent of the cab
accidents in the city involve green cabs and 15 percent involve blue cabs."
Now the information about base rates has been tied directly to the problem
at hand, and subjects do take it into account. It is too simple, I think, to
attribute subjects' apparent irrationality simply to the ways in which these
377
problems are phrased and the kinds of conversational postulates involved.

But at the same time, it is important not to ignore. the possible effects of
these factors (Kahneman, Slovic, and Tversky 1982, p. 157).
Jonathan Cohen uses the Tversky and Kahneman results as a means
of arguing that individuals are genuinely rational, and to suggest that
experimenters are caught up in statistical fine points having little bearing
upon survival in daily life. But other philosophers, far from questioning
the Tversky-Kahneman findings, have reformulated their news in light of
these results. Christopher Chemiak, for example, rejects the notion that
any rational agent must employ a logic, such as a complete first-order
deductive system (1983). Indeed, Chemiak suggests abandoning a quest
for ideal rationality: to achieve this goal, one would have to marshal
infinite resources. Chemiak instead suggests a standard of minimal ratio-
nality. One makes some but not necessarily all inferences that are appro-
priate, and even uses heuristics that are formally incorrect-sketched
images and other shorthand devices-as a means to arrive at a reasonable
decision within a reasonable time.
Henry Kyburg, philosopher from the University of Rochester, be-
lieves that philosophers are still entitled to describe rationality as an ideal
that individuals may only approach. But he goes on to indicate that empiri-
cal data are relevant in that they show how normative principles of infer-
ence, probability, and decision making are routinely violated by human
beings. Kyburg rejects Cohen's notion that the intuitions of ordinary peo-
ple should be the standards of rationality, and says, "It is not clear that
such an inquiry would be any more relevant to the development of norma-
tive standards of inductive or deductive logical cogency than an inquiry
into people's arithmetical intuitions would be to the development of stan-
dards of arithmetical validity" (Kyburg 1983, p. 232). But, Kyburg readily
acknowledges that "empirical studies may suggest certain facts relevant to
the development of normative constraints" (p. 244). Such a theory should
be based both upon normative factors and what people actually do. He
concludes:
Though I think it is clear that the problem of characterizing rationality is a

difficult one-far more difficult than many have realized-it does not seem insu-
perable. The very difficulties we uncover contribute to our understanding. And the
fact that we progress at all-the fact that we listen to each other's arguments and
recognize an obligation to deal with them-suggests that our goal can be ap-
proached. (P. 245)
Chemiak's and Kyburg's messages from the philosophical flank are

important for our inquiry. Rather than throwing away the issue of ratio-
nality altogether, or concluding that empirical work is strictly irrelevant,
378
they construe the study of rationality as a joint venture-one involving the
findings of empirical researchers as well as the clarifications and distinc-
tions of philosophical analysts. They leave open the possibility that in-
dividuals may depart from strict canons of logic and yet exhibit a viable
version of rationality. We see at work here a catholic model of cognitive
science. Instead of particular sciences banding together-as they have
done, say, in the study of psycholinguistics-there is a broader dialogue
at work-a dialogue between philosophers who are continuing to ponder
a classical philosophical problem, and empirical scientists who have de-
vised ingenious methods for attacking these venerable issues. And while
the examples here have been drawn largely from the realm of psychology,
they are by no means restricted to that discipline. Indeed, as I have sug-
gested, the kinds of concern explored by Wason and Johnson-Laird, by
Tversky and Kahneman, also involve issues of linguistics (How are the
particular questions phrased?); they can be applied to diverse anthropolog-
ical settings (How is reasoning carried out by different populations exhib-
iting different kinds of bias?); and they may be instructively simulated by
artificial intelligence (as Johnson-Laird has modeled his work on simula-
tion). Indeed, of the cognitive sciences, only the concerns of neuroscientists
stand relatively remote from work on logic, reasoning, and rationality.
Whether this statement is true only at present, or whether the whole realm
of rational and logical beliefs operates at a level apart from neuroscience,
has not yet been adequately considered.
Conclusion
With this review of work on rationality, I complete this survey of four

exemplary contemporary efforts in the cognitive sciences. Taken together,
these cases confirm that interdisciplinary efforts can achieve genuine prog-
ress in clarifying long-standing epistemological issues. While strong cri-
tiques have been leveled at this work-for example, the attack fashioned
by L. Jonathan Cohen-these can, in my view, be satisfactorily met with-
out fundamental damage to the cognitive-scientific enterprise. Such are-
sult is particularly impressive in an area like rationality, which is equally
remote from the operations of the nervous system and from the kinds of
elementary process involved in perceiving visual form.
The work on rationality, like that on categorization, yields an intrigu-
ing moral. Some decades ago, before the computer was invented and cogni-
tive science had been launched, it was common to maintain that human
379
beings typically form concepts of a classical sort and that they generally
reason in a logical fashion. Now the cognitive age, with its high-powered
computational techniques, has called into question the view of human
beings as operating in precise fashion. I do not mean, of course, that human
behavior is no longer subject to study by computational or other cognitive-
scientific techniques-indeed, Johnson-Laird has shown us just how some
such behavior can be accurately simulated; but the digital and deductive
fashion in which humans have been alleged to think is not viable. The
broader question remains whether various forms of human irrationality-
those documented by clinicians like Sigmund Freud and Carl Jung or by
anthropologists like Clifford Geertz and Dan Sperber-can be elucidated
by the methods of cognitive science.
Even negative lessons are important, and a science cannot be responsi-
ble for the message it yields, whether cheerful or gloomy. Yet challenging
the model of man-as-computer raises far-reaching questions of the extent
to which cognitive science has embraced the proper view of human menta-
tion and provided the proper methods for its study. I tum to these ques-
tions in the final chapter of the book, as there I revisit the major themes
of cognitive science in light of the histories I have related and the particular
lines of current research I have presented. In conclusion, I present my views
about the extent to which cognitive science has lived up to its initial
promise and delineate the principal paradoxes and challenges it confronts
at the present time.
380
14
Conclusion:
The Computational
Paradox and the
Cognitive Challenge
Surveying the scientific landscape at the beginning of the century, a far-
sighted observer might have felt justified in announcing the arrival of the
mind's new science. After all, building on the philosophical tradition of the
Greeks and the Enlightenment, and in the wake of dramatic breakthroughs
in physics, chemistry, and biology, the solution to the mystery of human
mental processes seemed at hand. Moreover, toward the end of the nine-
teenth century, a raft of new disciplines concerned particularly with
human thought and behavior had been launched. Surely the opportunity
to look at individuals in many cultures, in the light of the latest findings
about the human nervous system and with the powerful tools of logic and
mathematics, should sooner or later yield a bona fide science of the mind.
From a contemporary perspective, it seems evident that at least three
conditions had to fall into place before this dream could reach fruition.
First of all, it was necessary to demonstrate the inadequacies of the behav-
iorist approach. Second, the particular limitations of each social science had
to be acknowledged. Finally, the advent of the computer was needed to
provide the final impetus for a new cognitive science.
381
In the preceding chapters, I have shown how each of these three

conditions came to be met. By 1948, when Karl Lashley gave his famous
Hixon Symposium address on the problem of serial order in behavior, it
had become apparent to many scientists that the behaviorist approach to
human intellective activity was fatally flawed. By the same token, the
limits of other schools in the behaviorist orbit-logical positivism, struc-
tural linguistics, anthropological functionalism, Pavlovian reflexology-
were already becoming apparent. A fresh approach to these issues was
sorely needed.
Paralleling the discovery of the limitations of the behaviorist stance
was a growing realization that each of the several human and behavioral
sciences, practiced alone, harbored distinct and possibly crippling limita-
tions. Whether it was philosophy's ambivalenr.e about the relevance of
empirical data to long-standing epistemological issues, or psychology's
difficulty in adjusting its experimental approaches to large-scale issues, or
anthropology's problems in transcending the single case study, or neuro-
science's ambitions for dealing with capacities that defy reduction to the
neural level, these various sciences increasingly felt the need for fertiliza-
tion with neighboring disciplines.
Finally, and perhaps most decisively, there was the coalescence of
various mathematical and logical demonstrations (such as those of Shan-
non, Turing, and von Neumann) with important technological break-
throughs, which culminated around mid-century in the first computers.
Once the power of these machines for dealing with symbolic materials had
been demonstrated, many researchers became convinced that a science of
cognition might be fashioned in the image of the computer. By 1956,
psychologists such as George Miller and Jerome Bruner, computer scien-
tists such as Allen Newell and Herbert Simon, and linguists such as Noam
Chomsky had carried out work that (in retrospect) was cognitive-scientific
in spirit. And thirty years later, building on these pioneering efforts, re-
searchers such as David Marr and Stephen Kosslyn (working at the inter-
section of perceptual psychology and artificial intelligence), Eleanor Rosch
(combining psychological and anthropological concerns), and Philip John-
son-Laird (synthesizing approaches drawn from philosophy, psychology,
linguistics, and artificial intelligence) had demonstrated that clear progress
could be made in resolving long-standing philosophical and scientific is-
sues. Though work on the perceptual issues is further along than research
on classification or on rationality, it seems reasonable to declare in 1985
that cognitive science has come of age.
It is therefore opportune, in the life of the science as well as in the
course of this survey, to take stock: to revisit the principal themes of
cognitive science in order to clarify what has been accomplished over the
382
Conclusion: The ComputaHonal Paradox and the CogniHve Challenge
past few decades and to discern what remains to be accomplished if cogni-
tive science is to achieve its full potential. This evaluation will entail a
consideration of the central concept in cognitive science-that of the repre-
sentational level-as well as a re-examination of two themes introduced
in the opening chapters of this book, the computational paradox and the
cognitive challenge.
The Centrality of Mental RepresentaHon

To my mind, the major accomplishment of cognitive science has been
the clear demonstration of the validity of positing a level of mental repre-
sentation: a set of constructs that can be invoked for the explanation of
cognitive phenomena, ranging from visual perception to story comprehen-
sion. Where forty years ago, at the height of the behaviorist era, few
scientists dared to speak of schemas, images, rules, transformations, and
other mental structures and operations, these representational assumptions
and concepts are now taken for granted and permeate the cognitive
sciences.
While most researchers (and perhaps most readers) take the represen-
tational level for granted, this form of analysis must be situated with
reference to competing levels of description and analysis. It has long been
acceptable in empirical science to talk of the nervous system and, more
generally, of biological systems. These can, after all, be seen and even
dissected. While most physical scientists have been unconcerned profes-
sionally with cultural and historical matters, it has been acceptable (and
uncontroversial) among scholars in the humanities and social sciences to
offer explanations in terms of social forces, cultural practices, historical
traditions, and the like. How else, after all, to deal with macroscopic social
phenomena? The triumph of cognitivism has been to place talk of repre-
sentation on essentially equal footing with these entrenched modes of
discourse-with the neuronal level, on the one hand, and with the socio-
cultural level, on the other. Whoever wishes to banish the representational
level from scientific discourse would be compelled to explain language,
problem solving, classification, and the like strictly in terms of neurological
and cultural analysis. The discoveries of the last thirty years make such an
alternative most unpalatable.
Making the general case for representation is one thing, making it
with precision and power quite another. Any number of vocabularies and
conceptual frameworks have been constructed in an effort to characterize
the representational level-scripts, schemas, symbols, frames, images,
mental models, to name just a few. And any number of terms describe the
operations carried out upon these mental entities-transformations, con-
383
junctions, deletions, reversals, and so on. Cognitive science needs to put

its conceptual house in order and to transcend slogans and "buzz" words;
the field must agree upon a language for talking about a range of represen-
tational phenomena--even if that language turns out to harbor various
dialects.
As a start, I would single out two varieties of representation. One form
is initially or eventually built into the hardware-be it computer or brain.
Such a form must be invoked in order to detail what happens to informa-
tion but this variety of representation does not involve processes of which
the organism is in any way conscious or aware. For example, during the
early stages of visual processing described by Marr and his colleagues, the
visual system must create symbolic representations of physical informa-
tion and then operate on these representations. But no organism has any
options about these steps, and they are accessible only to a cognitive
scientist.
A second variety of representation encompasses those problem-solv-
ing and classificatory behaviors that individuals carry out with some flexi-
bility and some degree of explicitness and awareness. In analyzing a sen-
tence or a story, in creating an image or transforming it, one may well
become aware of having created some mental representation-or mental
model-and then one carries out operations upon that model. Explicit
awareness is not necessary here, but it is at least a possibility. Moreover,
the individual has the option of changing the mode of representation or
the kind of rule that is invoked. This mental activity is appropriately
described in terms of representational language but clearly warrants a
separate status (or terminology) from the kinds of representations that are
automatic and possibly wired in.
It may well be that there are several varieties of representation, or that
there exists a continuum from implicit to explicit, or from hard-wired to
flexibly programmed. But unless a taxonomy can be agreed upon, discus-
sion of representation will seem ad hoc and unsatisfactory. If representa-
tion is indeed the linchpin of cognitive science, it must ultimately be stated
as clearly and accepted as widely as quantum theory in physics or the
genetic code of the biochemical sciences. Such clarity and consensus seem
a long way off.
The ComputaHonal Paradox

Strictly speaking, one could have had cognitive science without the
computer. After all, computational theory antedated the invention of the
computer. And yet as a matter of historical fact, cognitive science was
unlikely to have arisen when it did, or taken the form that it has, without
384
Conclusion: The CompufaH.onal Paradox and the Cognitive Challenge
the emergence of the computer in our time. Since the first generation of
cognitive scientists, the computer has served as the most available and the
most appropriate model for thinking about thinking. And for most, it soon
became indispensable in their daily empirical and theoretical work.
Though the linking of computation and cognitivism turns out to have been
a contingent rather than a necessary fact, the fate of cognitive science is
closely tied to the fate of the computer.
And this leads to that strange state of affairs I have dubbed the
computational paradox. With the vigorous tradition, since the time of the
Greeks, of thinking about human thought as an embodiment of mathemat-
ical principles, it is hardly surprising that the first generation of cognitivists
-reared in the logical positivist tradition-should have embraced a highly
rationalistic view of human thought. One of the major results of the first
years of cognitive science, however, has been a challenge of that ready
assumption.
To be sure, when it comes to elementary and relatively "impenetra-
ble" processes like visual perception or syntactic analysis, an authoritative
computational account may some day be given. That is, the kinds of
descriptions that are legitimately offered in the terms of a digital von
Neumann computer may tum out to be appropriate accounts of these
human cognitive processes as well. But as one moves to more complex and
belief-tainted processes such as classification of ontological domains or
judgments concerning rival courses of action, the computational model
becomes less adequate. Human beings apparently do not approach these
tasks in a manner that can be characterized as logical or rational or that
entail step-by-step symbolic processing. Rather, they employ heuristics,
strategies, biases, images, and other vague and approximate approaches.
The kinds of symbol-manipulation models invoked by Newell, Simon, and
others in the first generation of cognitivists do not seem optimal for de-
scribing such central human capacities.
The paradox lies in the fact that these insights came about largely
through attempts to use computational models and modelingi only
through scrupulous adherence to computational thinking could scientists
discover the ways in which humans actually differ from the serial digital
computer-the von Neumann computer, the model that dominated the
thinking of the first generation of cognitive scientists.
I must again underscore one point. By insisting on the computational
paradox, I do not mean to assert that it is impossible to arrive at a computa-
tional account of human behavioral and thought patterns in all of their
perversity, irrationality, and subjectivity. Such accounts may well be pos-
sible and-as has been known since the time of Turing-are certainly
possible in principle. Rather, the paradox suggests that the portrait of
385
human cognition emerging from cognitive science is far removed-at least

at the molar level-from the orderly, precise, step-by-step image that
dominated the thinking of the founders of the field (and of those who
dreamed about it in the more distant past). Human thought emerges as
messy, intuitive, subject to subjective representations-not as pure and
immaculate calculation. These processes may ultimately be modeled by a
computer, but the end result will bear little resemblance to that view of
cognition canonically lurking in computationally inspired accounts.
Entailed in the reliance on the computer as a pivotal model of thought
is another difficulty which has only recently begun to be recognized. Invo-
cation of the computer leads naturally to a concentration on logical prob-
lem solving (d Ia Newell and Simon) or on orderly, highly rule-governed
analysis (d la Chomsky). But evidence from neuropsychological and devel-
opmental studies of mental processes has indicated that our concepts of
cognition need to be considerably broadened. Processes involved in musi-
cal and other artistic activities, and, quite possibly, processes involved in
knowing other individuals and in knowing oneself merit the modifier
cognitive (Gardner 1983). To the extent that this position is valid, a thor-
oughgoing cognitive science will need to account for these abilities as well
as for more familiar logical mathematical applications of mind. Whatever
their relevance for the study of human rationality or problem solving,
models derived from the computer are even less likely to be adequate to
account for these other uses of mind.
It could be countered that cognitive science ought to be satisfied with
modeling logical thought and that these other forms of thought ought to
fall by the wayside. Perhaps cognitive science should embrace a classical
computational account, even if humans do not much resemble a classical
kind of computer. To restrict cognitive science to one form of cognition,
however, is to refashion the subject matter to fit the current tools of study.
By the same token, to accept an account just because such an account can
be given is a scientifically weak move. After all, the purpose of science is
not to propose a possible analysis (of which there will always be an infinite
number) but rather to come up with the analysis that is most appropriate,
parsimonious, and convincing. All cognitive phenomena could, after all, be
described in terms of atoms, or in terms of historical factors-and in either
event, a representational account would not even be necessary. But now,
at the very time when a representational account has been accepted, it is
important to try to find the optimal representational account. Representa-
tion without computation is one possible outcome for certain regions of
cognitive science.
The idea of representation has until this point been closely tied to
our current conceptions of computers. But there is no way of determin-
386
Conclusion: The Computational Paradox and the Cognitive Challenge
ing a priori to what extent the ways currently embraced for describing the
representations of computers will prove germane to organisms, be they
paramecia or professors. The kinds of representations favored by neo-
associationists like Geoffrey Hinton tum out to be radically different
(and much sparer) than those countenanced by Jerry Fodor or Zenon
Pylyshyn; moreover, it may turn out that neither is adequate or suited
for describing an individual who is dreaming, writing a poem, or listen-
ing to music. Earlier models of thought-the reflex arc, the hydraulic
engine, the telephone switchboard-are now seen to be extremely lim-
ited. It is already clear that one kind of computer does not suffice to
model all thought. We must face the alternative that humans may be an
amalgam of several kinds of computers, or computer models, or may de-
viate from any kind of computer yet described. Computers will be pivo-
tal in helping us determine how computerlike we are, but the ultimate
verdict may be "Not very much."
Even if computers emerge as viable models for certain facets of human
thought, the question arises about the various aspects of human nature
that have been bracketed by cognitive scientists. As I noted in chapter 3,
nearly all cognitive scientists have conspired to exclude from consideration
such nontrivial factors as the role of the surrounding context, the affective
aspects of experience, and the effects of cultural and historical factors on
human behavior and thought (see D. Norman 1980). Some take the posi-
tion that this is only a temporary move, until the relatively discrete aspects
of cognition have been unraveled; others take the stronger positions that
cognitive science should never deal with these aspects or even that a
cognitive-scientific account will ultimately render unnecessary any ac-
count of these "fuzzier" factors.
Even a brief consideration of each of these "bracketed" topics would
require many pages, and since cognitive scientists have themselves steered
clear of these issues, there is little work within the disciplinary tradition
on which I can draw. My own belief is that, ultimately, cognitive science
will have to deal with these factors in one of two ways. Either scientists
will propose a cognitive account of affect in which, for example, affective
states will be viewed as quantitative values along a dimension, like happi-
ness or cruelty; or researchers will opt for a complex explanatory frame-
work in which the interaction of traditional cognitive factors with affective
or cultural factors can somehow be modeled. These will be important but
enormously difficult undertakings, for which traditional computational
considerations may provide scant help.
Scholars differ widely from one another in their intuitions about the
extent to which these other factors may ultimately engulf cognitive fac-
tors. From the perspective of a philosopher like Hubert Dreyfus, a linguist
387
like Roy Harris, a psychologist like Benny Shanon, or an anthropologist

like Clifford Geertz, these factors are so important, so constitutive of
human experience, that they, rather than cognitive factors, ought to be
regarded as primary. Although I am not without sympathy for this per-
spective, my own view is that there is a heartland of cognition which can
be accounted for on its own terms, without necessary reference to (or
reliance upon) these other, undoubtedly important elements. It is this
heartland that I have attempted to describe in this book, particularly in the
last four chapters. The borders of this heartland may determine the limits
My own doubts about the computer as the guiding model of human
thought stem from two principal considerations. As Hilary Putnam (1981)
has stressed, the community surrounding a cognizing individual is critical.
From those around us, we come to understand which sorts of views are
considered acceptable, which are false or dangerous, justified or unjus-
tified. Such judgments cannot initially be made by an individual but must
stem from a collectivity; and because we all belong to communities, it
makes sense for us to invoke such judgments. In sharp contrast, it makes
no sense to indicate that a computer has made a mistake or is unjustified
in its beliefs. The computer is simply executing what it has been pro-
grammed to execute, and standards of right and wrong do not enter into
its performance. Only those entities that exist within, interact with, and
are considered part of a community can be so judged.
My other reservation about the computer as model centers on the deep
difference between biological and mechanical systems. I find it distorted
to conceive of human beings apart from their membership in a species that
has evolved over the millennia, and as other than organisms who them-
selves develop according to a complex interaction between genetic pro-
clivities and environmental processes over a lifetime. To the extent that
thought processes reflect these bio-developmental factors and are suffused
with regressions, anticipations, frustrations, and ambivalent feelings, they
will differ in fundamental ways from those exhibited by a nonorganic
system. Note that it did not have to be this way-biological systems might
have been just like inorganic (mechanical) systems. But it is clear that they
are not. I therefore believe that adequate models of human thought and
behavior will have to incorporate aspects o( biological systems (for exam-
ple, processes of organic differentiation or fusion) as well as aspects of
mechanical systems (the operation of electronic circuits). The very com-
parisons between organic and mechanical structures and processes may be
among the most instructive aspects of the science. All told, cognitive
science will have to incorporate (and connect to) neurobiology as much as
to artificial intelligence.
388
The Cognitive Challenge
Central to my view of cognitive science is the claim that the field

entails an empirical effort to answer long-standing epistemological ques-
tions. Classical philosophy has indeed supplied much of the intellectual
agenda of the contemporary field: its interest in how we perceive the
world, in how we classify objects, in the status of words, images, and other
constructs, and in the assessment of human rationality or irrationality.
And even those issues that could not have been formulated by the classical
philosophers-such as the extent to which human thought is computa-
tional-have routinely been put forth in philosophical terms. Perhaps most
notably, contemporary cognitive science has provided reasonable answers
to selected philosophical questions, even as it has rejected certain issues
and radically transformed others.
In a sense, philosophy can be seen as standing outside of mainstream
empirical cognitive science. On one bank of the mainstream, philosophy
supplies many of the issues to be investigated. On the other, it examines
the answers that are forthcoming, helps to interpret and integrate them,
and provides critiques of the overall enterprise. Thus, for example,
philosophers first raised the issues of human rationality and have now
participated vigorously and instructively in the interpretation of findings
put forth by Freud, Piaget, Tversky, and Kahneman. Yet as members of a
discipline that stands external to empirical science, philosophers concerned
with cognitive science may seem in jeopardy. For, as philosophical ques-
tions are answered by empirical science, philosophers may ultimately re-
cede from the scene-as has, in fact, happened in vast areas of physics and
biology. Still, in my view, philosophers interested in cognition stand in no
peril of intellectual unemployment for the foreseeable future.
But what of the relations obtaining among the other disciplines that
make up the cognitive sciences? Are they likely to blend together into one
seamless Cognitive Science, or can we expect them to maintain their au-
tonomy in the years ahead? And what would be the most favorable state
of affairs?
One can contrast two visions of cognitive science. The less ambi-
tious one calls for cooperation among the six member disciplines, each
still retaining its primary questions, methods, and goals as chronicled in
part II. On such an account, philosophy supplies the principal issues and
helps to judge the extent to which they have been successfully handled.
Neuroscience and anthropology remain as border disciplines, psychology
and artificial intelligence are the core disciplines, and linguistics offers
an account of that ability which is most central in the human cogni-
tive armamentarium. When collaborating, these researchers are "practic-
389
ing cognitive scientists"; otherwise, they are simply doing their own
thing.
This "weak" version of cognitive science is quite possibly the norm
today but scarcely warrants the label of an important new science. In a
stronger, more gritty version of cognitive science, there will be gradual
attenuation of disciplinary boundaries and loyalties. These will be replaced
by a concerted effort by scientists committed to a representational account
to model and explain the most crucial human cognitive functions.
This reconfiguration of the territory of cognitive science rests on the
following analysis. Today, what is most central to cognitive studies is an
individual's disciplinary background: whether one works as a philosopher
or an anthropologist is more salient than whether one works on issues of
language or of social interaction. This organization around the traditional
disciplines would be appropriate if the actual domains of cognition did not
make a central difference; so long as the same processes are believed to
occur irrespective of the content of a domain (musical versus spatial cogni-
tion, for example), the conventional disciplinary division of labor makes
sense.
I hold a very different, and still controversial, vision. From my per-
spective, as elaborated in this book, the crucial divisions within cognitive
science are not the traditional disciplinary perspectives but rather the spe-
cific cognitive contents. Therefore, scientists should be characterized by
the central cognitive domain on which they work: broad domains like
language, music, social knowledge, logical thought; and more focused sub-
domains, like syntactic processing, the early phases of visual processing,
or the perception of rhythm. Scientific training and research enterprises
should come increasingly to be organized around these problems. When
working on these problems, scientists should fuse their necessarily differ-
ent perspectives in order to arrive at a full account of the particular cogni-
tive domain at issue. And so the ultimate cognitive-scientific picture of
syntactic processing, or of language as a whole, should be a coordinated
representational account which covers the full gamut of the traditional
disciplines without any need even to mention them.
Yet the question of disciplines or, more broadly, of levels of explana-
tion cannot be bypassed entirely-and here we confront the major chal-
lenge to contemporary cognitive studies. Having established the legitimacy
of the representational level, cognitive workers must trace out the ways in
which this level maps onto the other legitimate (and legitimized) ways in
which human activities can be construed. For a time, believers in the
representational level had to proceed along their own, as yet unexplored
path-and adherence to this single-minded program was the genius of the
pioneering generation of the 1950s. But ultimately, such splendid isolation
must be shattered. We must come to understand how culture is mapped
390
onto brains-and the royal road toward such understanding will be the
representational level.
The reason for such linkage across levels is simple but crucial. Unless
the significance of work in each science can be connected to that under-
taken in neighboring areas, the significance (and the limitations) of that
work cannot be appreciated. No one fears the demise of physics, chemis-
try, and biology; and yet each of these discipines has vital, articulated,
necessary links to the next level, through "borderland" disciplines like
physical chemistry or biochemistry.
But, paradoxically, much of the best work in cognitive science has
been carried out as if only the level of mental representation existed. In the
case of language (more specifically, grammar), for example, the brilliant
work of Chomsky and his followers makes no reference to, and could be
maintained irrespective of, the actual conditions in the brain and in the
surrounding culture. If cognitive science is to mature, however, the ulti-
mate representational account of language must relate, at one extreme, to
knowledge about the neural architecture of certain regions of the left
hemisphere of the brain; and, on the other, to knowledge about the struc-
ture and function of language in different cultural groups. Only such a
linking of levels can indicate whether proposed representational accounts
of language are in fact appropriate, in light of neural and cultural consider-
ations. The goal of this penetration of levels is not, to repeat, so that one
discipline or level can swallow the other; but rather, so that our under-
standing of a domain like language can touch on all the relevant scientific
perspectives, from neuron to nation.
Just as many committed cognitive scientists have restricted their work
to the representational level and have spumed the borderland territory, so
too they have called for a narrow delineation of what counts as cognitive.
Thus, Jerry Fodor (1981) has expressed skepticism about the capacity of
cognitive science to explain any of the higher or inore complex forms of
thought, which are "permeable" to a person's beliefs; and Zenon Pylyshyn
(1984) has proposed a definition of cognition which excludes areas like
learning, development, and "moods."
Although perhaps a prudent research strategy in the short term, I find
this a misguided overall program for the cognitive sciences. Just because
our current measures or concepts are primitive, we ought not to violate a
common-sense notion of what mind is about; even less should we want
to bypass the most impressive achievements of the human mind.
Indeed, in my view, the ultimate goal of cognitive science should be
-precisely-to provide a cogent scientific account of how human beings
achieve their most remarkable symbolic products: how we come to com-
pose symphonies, write poems, invent machines (including computers), or
construct theories (including cognitive-scientific ones). Such accounts will
391
I I I I TOWARD AN INTEGRA TED CoGNITIVE SciENCE
have to incorporate the means by which humans embark on complex

projects to achieve ambitious goals; how they represent their plans; how
they initiate work on a project, organize their daily routines (and nonrou-
tines!), evaluate tentative drafts in light of feedback from other people and
in view of their own motives and standards, determine when such a pro-
gram or product has been completed, and then initiate a new line of work.
Such an exploratory enterprise will probably entail mentalistic entities
or models of a highly molar form (as well as many finer-grained entities).
It will probably cut across narrow domains (like language-related pro-
cesses) and also have to involve separate constructs to account for pro-
cesses involved in creativity, synthesis, and/or consciousness. Ultimately,
as part of the cognitive challenge, it will also be necessary to relate a
representational account of these human intellectual achievements to what
is known about their neural substrate and to what can be established about
the role of the surrounding culture in sponsoring and then absorbing (or
rejecting or refashioning) them.
Even to begin to outline the phases involved in modeling any complex
human creative activity is to confront the immensity of the task and the
primitiveness of our current tools. And yet it is crucial for cognitive scien-
tists to keep this goal in mind-even to tack it over their desks or alongside
the screens of their personal computers. The study of thought must not
exclude its most remarkable exemplars even if their elucidation still seems
remote.
Given the most optimistic scenario for the future of the cognitive
sciences, we still cannot reasonably expect an explanation of mind which
lays to rest all extant scientific and epistemological problems. Still, I believe
that the authors of Meno, The Discourse on Method, The Critique of Pure Reason,
and The Origin of Species would feel that distinct progress has been made on
the age-old issues that exercised them. Thanks to the development of new
logical tools, the diverse deployments of the computer, the application of
the scientific method to human psychological processes and cultural prac-
tices, our deeper and more rigorous understanding of the nature of lan-
guage, and the many discoveries about the organization and operation of
the nervous system, we have attained a more sophisticated grasp on the
issues put forth originally by Plato, Descartes, Kant, and Darwin.
How much further cognitive science can proceed, and which of the
competing visions it will choose to pursue, are issues that remain open. All
who style ourselves as cognitive scientists are on the spot. If we heed the
lessons entailed in our scientific history and lurking in our philosophical
backgrounds, if we attend to but are not stymied by the reservations aired
by shrewd skeptics, if we recognize the limitations of all inquiry but
do not thereby encounter a failure of nerve, there are clear grounds for
optimism.
392
REFERENCES
Abel, R. 1976. MRn the MeRSure. New York: Free Press.

Adler,]. E. 1984. "Abstraction Is Uncooperative," fournRI for the Theory of SodRI BehRvior 14:
165-181.
Allport, D. A. 1980. "Patterns and Actions: Cognitive Mechanisms Are Content-specific." In
G. L. Oaxton, ed., Cognitive Psychology: New Directions. London: Routledge & Kegan Paul.
Anderson, A. R., ed. 1964. Minds Rnd MRchines. Englewood Cliffs, N.J.: Prentice-Hall.
Anderson,]. R. 1978. "Arguments concerning Representations for Mental Imagery," Psycholog-
icAl Review 85: 249-77.
- - - . 1980. Cognitive Psychology Rnd Its lmplicRtions. San Francisco: W. H. Freeman.
- - - . 1983. The Architecture of Cognition. Cambridge, Mass.: Harvard University Press.
Armstrong, S. L.; Gleitman L. R.; and Gleitman, H. 1983. "What Some Concepts Might Not
Be," Cognition 13: 26.3-308.
Arnheim, R. 1969. VJSuRl Thinking. Berkeley: University of California Press.
Ashby, W. Ross. 1952. Design for a BrRin. New York: John Wiley.
Atkinson, R. L.; and Shiffrin, R. M. 1968. "Human Memory: A Proposed System and Its
Control Processes." InK. W. Spence and]. T. Spence, eds., The Psychology of LeArning and
MotiVAtion: AdVAnces in ReseArch Rnd Theory, vol. 2. New York: Academic Press.
Austin, J. L. 1962. How to Do Thing5 with Words. Cambridge, Mass.: Harvard University Press.
Ayer, A. J. 1936. LRnguRge, Truth, Rnd Logic. New York: Dover.
- - - . 1973. The CenfrRl Questions of Philosophy. Harmondsworth, Middlesex, England: Penguin
Books.
- - - . 1982. Philosophy in the Twentieth Century. New York: Random House.
Baddeley, A. 1981. "The Cognitive Psychology of Everyday Life," British fournRl of Psychol-
ogy 72: 257-69.
Bahrick, H. P.; Bahrick, P. 0.; and Wittlinger, R. P. 1975. "Fifty Years of Memory for Names
and Faces: A Cross-sectional Approach," fournRl of Experimental Psychology: GenerRl 104:
54-75.
Ballard, D. H.; Hinton, G. E.; and Sejnowski, T. ]. 1983. "Parallel Visual Computation,"
NRture 306: 21-26.
Barnes, J. 1979. The PresocrRtic Philosophers. London: Routledge & Kegan Paul.
Bartlett, F. C. 1932. Remembering. Cambridge: Cambridge University Press.
- - - . 1958. Thinking: An ErperimeniRI and SociRI Study. London: George Allen & Unwin.
Bateson, G.; Jackson, D. D.; Haley, ].; and Weakland, J. 1956. "Toward a Theory of Schizo-
phrenia," Behavioral Science 1: 251-64.
Beach, F. A.; Hebb, D. 0.; Morgan, C. T.; and Nissen, H. W., eds. 1960. The Neuropsychology
of LRShley. New York: McGraw-Hill.
Beaumont, J. C.; Young, A. W.; and McManus, J. C. 1984. "Hemisphericity: A Critical
Review," Cognitive Neuropsychology 1: 191-212.
Benedict, R. 1934. PRtterns of Culture. Boston: Houghton-Mifflin.
- - - . 1946. The ChrySAnthemum Rnd the Sword: PRfferns of japAnese Culture. Boston: Houghton-
Mifflin.
Berkeley, G. 1929. EsSRys, Principles, DiAlogues with Selections from Other Wrifing5. M. W. Calkins,
ed. New York: Charles Scribner.
Berlin, B.; Breedlove, D. E.; and Raven, D. H. 1973. "General Principles of Classification and
Nomenclature in Folk Biology," American Anthrupology 75: 214-42.
393
References
Berlin, B.; and Kay, P. 1969. BflSic Color Terms: Their UniversRiity Rnd Evolution. Berkeley: Univer-
sity of California Press.
Bernstein, J. 1981. "Profiles: Marvin Minsky and Artificial Intelligence," New Yorker 57:
50-126.
- - - . 1982. Science Observed: EsSRys Out of My Mind. New York: Basic Books.
Berwick, R.; and Weinberg, A. 1983. "The Role of Grammars in Models of Language Use,"
Cognition 13: 1-61.
Block, N., ed. 1981R. lmRgery. Cambridge, Mass.: MIT Press, Bradford Books.
- - - , ed. 1981b. Readings in Philosophy of Psychology, vol. II. Cambridge, Mass.: Harvard Univer-
sity Press.
- - - . 1983. "Mental Pictures and Cognitive Science," The PhilosophicAl Rroiew 92: 499-541.
Bloomfield, L. 1925. "Why a Linguistics Society?," LRnguRge 1: 1-5.
- - - . 1933. LRnguage. 2nd. ed. New York: Holt, Rinehart & Winston.
- - - . 1936. "Language or Ideas?," LRnguage 12: 89-95.
- - - . 1943. "Obituary of Franz Boas," LRnguage 19: 198.
Blumenthal, A. L. 1979. "Psychology and Linguistics: The First Half-century." Lecture deliv-
ered at the APA centennial symposium: "Linguistics and Psychology," New York.
Bobrow, D. G. 1968. "Natural Language Input for a Computer Problem-solving System." In
M. L. Minsky, ed., SemRntic Information Processing. Cambridge, Mass.: MIT Press.
---;and Winograd, T. 1977. "An Overview of KRL, a Knowledge Representation Lan-
guage," Cogniti'OI! Science 1: 3--46.
Boden, M. 1977. Arti/iciRllntelligence and NRiurRI MRn. New York: Basic Books.
- - - . 1981. Minds Rnd MechRnisms. Ithaca, N.Y.: Cornell University Press.
Boffey, P.M. 1983. "'Rational' Decisions Prove Not to Be," New York Times, 6 December 1983,
pp. Cl, C7.
Bolinger, D. 1968. Aspects of LRnguRge. New York: Harcourt, Brace & World.
Boring, E. G. 1950. A History of ExperimentAl Psychology. New York: Appleton-Century-Crofts.
Bousfield, J. 1979. "The World Seen as a Colour Chart." In R. F. Ellen and D. Reason, eds.,
ClRssi/icRtions in Their SocUll Context. London: Academic Press.
Bower, G. 1976. "Comprehending and Recalling Stories." Address to the American Psycho-
logical Association, Washington, D.C.
Brainerd, C. J. 1978. "The Stage Question in Cognitive-Developmental Theory," The Behavioral
Rnd BrRin Sciences 2: 173-213.
Bransford, J. D.; and Franks, ]. J. 1971. "The Abstraction of Linguistic Ideas," Cogniti'OI!
Psychology 2: 331-50.
Bransford, J. D.; Franks, J. J.; Morris, C. D.; and Stein, B. S. 1979. "Some Constraints on
Learning and Memory Research." In L. S. Cermak and F. I. M. Craik, eds., Lroels of
Processing Rnd HumRn Memory. Hillsdale, N.J.: Erlbaum.
Bransford, }. D.; and Johnson, M. V. 1972. "Contextual Prerequisites for Understanding:
Some Investigations of Comprehension and Recall," Journal of VerbRI Learning Rnd VerbRI
BehRvior 11: 717-21.
Bransford, J.D.; and McCarrell, N. S. 1975. "A Sketch of a Cognitive Approach to Compre-
hension: Some Thoughts about Understanding What It Means to Comprehend." In P.
N. Johnson-Laird and P. C. Wason, eds., 1977. Thinl:i.ng: Readings in Cognitive Science. Cam-
bridge: Cambridge University Press.
Brentano, F. 1874. Psychologie vom empirischen StRndpunkte. Leipzig.
Bresnan, J, 1978. "A Realistic Transformational Grammar." In M. Halle, et al., eds., Linguistic
Theory Rnd PsychologicRI ReAlity. Cambridge, Mass.: MIT Press.
- - - . 1981. "An Approach to Universal Grammar and the Mental Representation of Lan-
guage," Cognition 10: 39-52.
Brew, J. 0., ed. 1968. One Hundred Years of Anthropology. Cambridge, Mass.: Harvard University
Press.
Brewster, D. 1854. North British Rroiew. Quoted in J. Miller, 1983. States of Mind. New York:
Pantheon Books.
Broadbent, D. E. 1954. "The Role of Auditory Localization in Attention and Memory Span,"
fournRI of ExperimentAl Psychology 47: 191-96.
- - - . 1980. "The Minimization of Models." In A.}. Chapman and D. M. Jones, eds., Models
of MRn, pp. 113-27. London: British Psychological Society.
Broca, P. 1861. "Remarques sur Ia siege de la faculte du langage articule," Bulletin de Ia societe
d'Rnlhropologie (Paris) 6.
394
References
Brown, C. M. 1984. "Computer Vision and Natural Constraints," Science 224: 1299-1305.
Brown, R. 1956. "Language and Categories." In J. S. Bruner et al., A Study of Thinking. New
York: John Wiley.
- - - . 1975. "Reference: In Memorial Tribute to Eric Lenneberg," Cognition 4: 125-53.
- - - ; and Lenneberg, E. H. 1954. "A Study in Language and Cognition," Journal of Abnormal
and Social Psychology 44: 454--62.
Bruner, J. S. 1944. Mandate from the People. New York: Duell, Sloan & Pearce.
- - - . 1973. Beyond the Information Given; Studies in the Psychology of Knowing. ]. Anglin, ed. New
York: W. W. Norton.
- - - . 1982. In Search of Mind. Preliminary version.
- - - . 1983. In Search of Mind. New York: Harper & Row.
--;Goodnow,].; and Austin, G. 1956. A Study of Thinking. New York: John Wiley.
Carnap, R. 1959. "Psychology in Physical Language." In A. J. Ayer, ed., Logical Positivism.
Glencoe, Ill.: Free Press.
- - - . 1967. The Logical Structure of the World; Pseudoproblems in Philosophy. Berkeley: University
of California Press. Original work published in 1928.
Cazeneuve, J. 1968. "Levy-Bruhl, Lucien." Entry in International Encyclopedia of the Social
Sciences, ed. by D. Sills. New York: Macmillan.
- - - . 1972. Luden Uvy-Bruhl, trans. by P. Riviere. Oxford: Basil Blackwell.
Chemiak, C. 1983. "The 'universal acceptance of logic.'" Berkeley Cogniti"OI! Science Report, no. 6.
Cherry, E. C. 1953. "Some Experiments on the Recognition of Speech, with One and with
Two Ears," Journal of the Acoustical Sodety of America 25: 975-79.
Chomsky, N. 1955. "The Logical Structure of Linguistic Theory." Doctoral dissertation,
University of Pennsylvania. Published as a monograph by Plenum Press, New York, in
1975.
- - - . 1957. SyntRCtic Structures. The Hague: Mouton.
- - - . 1964R. "A Review of B. F. Skinner's Verbal Behavior." In J. A. Fodor and J. J. Katz, eds.,
The Structure of LRnguRge: Readings in the Philosophy of LRnguRge, pp. 547-78. Englewood Cliffs,
N.J.: Prentice-Hall. Original work published in Language 35 (1959): 26-58.
- - - . 1964b. "A Transformational Approach to Syntax." In J. A. Fodor and J. }. Katz, eds.,
The Structure of LRnguRge: ReRdings in the Philosophy of Language, 1964, pp. 211-45. (Reprinted
from A. A. Hill, ed. (1962). Proceedings of the Third TaRS Conference on Problems of Linguistic
Analysis in English, 1958, pp. 124-58. Austin, Tex.: University of Texas Press.) Englewood
Cliffs, N.J.: Prentice-Hall.
- - - . 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press.
- - - . 1966. Cartesian Linguistics. New York: Harper & Row.
- - - . 1972. LanguRge and Mind. New York: Harcourt Brace Jovanovich.
- - - . 1975. Reflections on LRnguage. New York: Pantheon.
- - - . 1979. Language and Responsibility. New York: Pantheon. Originally published in French,
1977.
- - - . 1980. Rules and Representations. New York: Columbia University Press.
- - - . 1981. "Knowledge of Language: Its Elements and Origins," Philosuphical TransRCtions of
the Royal Society of London, Series B, 295: 223-34.
- - - . 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge,
Mass.: MIT Press.
- - - . 1984. "Changing Perspectives on Knowledge and Use of Language." Paper delivered
at the Society for Philosophy and Psychology, Massachusetts Institute of Technology,
Cambridge.
---;and Halle, M. 1968. The Sound Pattern of English. New York: Harper & Row.
Clowes, M. 1971. "On Seeing Things," Arti/icial Intelligence 2: 79-116.
Cohen, G. 1977. The Psychology of Cognition. London, New York: Academic Press.
Cohen, L. J. 1981. "Can Human Irrationality Be Experimentally Demonstrated?," Behavioral
and Brain Sciences 4: 317-70.
Colby, K. M. 1975. Artificial Paranoia. New York: Pergamon Press.
Cole, M.; and Scribner, S. 1974. Culture and Thought. New York: John Wiley.
Collins, A. 1977. "Why Cognitive Science," Cogniti"OI! Science 1: 3-4.
Copleston, F. 1964. A HIStory of Philosophy: Modern Philosophy: The British Philosophers. New York:
Doubleday.
Craik, F. I. M.; and Lockhart, R. S. 1972. "Levels of Processing: A Framework for Memory
Research," Journal of Verbal leRrning and Verbal Behavior 11: 671-84.
395
References
Crick, M. 1976, Explorations in LRnguage and MNning: ToWRrds a Semantic Anthropology. New York:
John Wiley.
D' Andrade, R. G. 1981. "The Cultural Part of Cognition," Cognitive Science 5: 179-95.
- - - . 1982. "Reason versus Logic." The Ecology of Cognition: Biological, Cultural and Historical
Perspectives. Symposium, Greensboro, N.C.
Dasen, P.R. 1972. "Cross-Cultural Piagetian Research: A Summary," jOJtmRl of Cross-Cultural
---;and Heron, A. 1981. "Cross-Cultural Tests of Piaget's Theory." In H. C. Triandis and
A. Heron, eds., Handbook of Cross-Cultural Psychology, vol. 4: Developmenflll Psychology. Boston:
Allyn & Bacon.
Davis, M. 1958. Computability and UnsolMbility. New York: McGraw-Hill.
Dejerine, J. 1892. "Contribution a !'etude anatomo-pathologique et clinique des differents
varietes de cecite verbale." Comp. rend. sceRn. soc. bioi. 9( 4): 61-90.
Dennett, D. C. 1978. Brainstorms: Philosophical Es5Rys on Mind and Psychology. Cambridge, Mass.:
MIT/Bradford Books.
- - . 1983. "Cognitive Wheels: The Frame Problem of A.I." Unpublished paper, Tufts
University.
Dennis, M. 1980. "Language Acquisition in a Single Hemisphere: Semantic Organization."
In D. Caplan, ed., Bwlogical Studies of Mental Processes. Cambridge, Mass.: MIT Press.
- - - ; and Whitaker, H. A. 1976. "Language Acquisition Following Hemidecortication:
Linguistic Superiority of the Left over the Right Hemisphere," Brain and LRnguage 3:
404-33.
Descartes, R. 1951. Meditation on First Philosophy, trans. by L. J. Lafleur. New York: Library on
Liberal Arts: Liberal Arts Press. Original work published in 1641.
de Valois, R. L.; and Jacobs, G. H. 1968. "Primate Color Vision," Science 162: 533-40.
De Waal Malefijt, A. 1974./mt~ges of Man: A History of Anthropologic11l Thought. New York: Alfred
A. Knopf.
Diamond, S., ed. 1974. The Roots of Psychology. New York: Basic Books.
Dodwell, P. 1971. "Is a Theory of Conceptual Development Necessary?" InT. Mischel, ed.,
Cognitive Development 11nd Epistemology. New York: Academic Press.
s
Donaldson, M. 1978. Children Minds. London: Fontana.
Danders, F. C. 1868. "On the Speed of Mental Processes." Reprinted in Ada Psychologica 30
(1969): 412-31.
Dorf, R. C. 1974. Computers 11nd M11n. San Francisco: Boyd & Frazer.
Dresher, B. E.; and Hornstein, N. 1976. "On Some Supposed Contributions of Artificial
Intelligence to the Scientific Study of Language," Cognition 4: 321-48.
Dreyfus, H. 1972. What Computers Can't Do: A Critique of Arti/icilll ReRSon. New York: Harper &
Row.
Duncker, K. 1945. "On Problem Solving," Psychological Monographs 58 (whole no. 270): 5.
Durant, W. 1926. The Story of Philosophy. New York: Simon & Schuster.
Durbin, M. 1966. "The Goals of Ethnoscience," Anthropological LingJtistics 8: 22-33.
Ebbinghaus, H. 1913. Memory: A Contribution fo Experimental Psychology. New York: Columbia
Teacher's College. Originally published 1885.
Edgerton, R. B.; and Langness, L. L. 1974. Methods 11nd Styles in the Study of Culture. New York:
Thomas Crowell.
Ehrenfels, C. von. 1890. "Uber Gestaltqualitaten." Vtijsch. Wrss. Philos. 14: 249-92.
Evans, T. G. 1968. "A Program for the Solution of Geometric-Analogy Intelligence Test
Questions." In M. L. Minsky, ed., Semantic lnfarmation Processing. Cambridge, Mass.: MIT
Press.
Fahlman, S.C.; Hinton, G. E.; and Sejnowski, T. J. 1983. "Massively Parallel Architecture for
AI; NETI, Thistle and Boltzmann Machines." Unpublished manuscript, Carnegie-Mel-
lon University.
Fancher, R. E. 1979. Pioneers of Psychology. New York: W. W. Norton.
Farah, M. J. 1984. "The Neurological Basis of Mental Imagery; A Componentical Analysis,"
Cognition 18: 243-61.
Fechner, G. T. 1912. "Elemente der psychophysik." In B. Rand, ed., Classical Psychologists, trans.
by H. S. Langfeld. Boston: Houghton-Mifflin. Original work published in 1860.
Feigenbaum, E. A. 1959. "An Information-Processing Theory of Verbal Learning." Doctoral
dissertation, Pittsburgh: Carnegie Institute of Technology, 1959. Rand Corporation
Paper, P-1817.
396
References
---;Buchanan, B. G.; and Lederberg, J. 1971. "On Generality and Problem Solving: A Case
Study Using the DENDRAL Program." In B. Meltzer and D. Michie, eds., Machine
Intelligence, vol. 6. Edinburgh: Edinburgh University Press.
Feigenbaum, E. A.; and Feldman, J., eds. 1963. Comflltters and Thought. New York: McGraw-Hill.
Feigenbaum, E. A.; and McCorduck, P. 1983. The Fifth Genm1tion: Artificial Intelligence and Japans
Comflltter Challenge to the World. Reading, Mass.: Addison-Wesley.
Feldman, J. A. 1981. "A Connectionist Model of Visual Memory." In G. E. Hinton and J. A.
Anderson, eds., Parallel Models of Assodati'OI! Memory. Hillsdale, N.J.: Lawrence Erlbaum.
Finke, R. A.; and Schmidt, M. J, 1978. "The Quantitative Measure of Pattern Representation
in Images using Orientation-specific Color After-effects," Perception and Psychophysics 23:
515-20.
Fodor, ]. A. 1972. "Some Reflections on L. S. Vygotsky's 'Thought and Language,'" Cog-
nition 7: 83-95.
- - - . 1975. The language of Thought. New York: Thomas Y. Crowell.
- - - . 1980. "Methodological Solipsism Considered as a Research Strategy in Cognitive
Psychology," Behavioral and Brain Scienm 3: 63-110.
- - . 1981a. "The Mind-Body Problem," Scientific American 244: 114-23.
- - - . 1981b. Representations: Philosophical Essays on the Foundations of Cogniti'OI! Science. Cambridge,
Mass.: MIT Press.
- - - . 1983. The Modularity of Mind. Cambridge, Mass.: MIT/Bradford Press.
- - - . 1984. "Information and Association." Unpublished paper, Massachusetts Institute of
Technology.
---;Bever, T. G.; and Garrett, M. 1974. The Psychology of language. New York: McGraw-Hill.
Fodor, J. A.; and Katz, J. J., eds. 1964. The Structure of Language: Readings in the Philosophy of
Language. Englewood Oiffs, N.J.: Prentice-Hall.
Fodor, J. A.; and Pylyshyn, Z. W. 1981. "How Direct Is Visual Perception: Some Reflections
on Gibson's 'Ecological Approach,'" Cognition 9: 139-96.
Foster, J. M. 1967. List Processing. London: Macdonald.
Foucault, M. 1970. The Orrkr of Things. New York: Pantheon.
Fox, J, K. 1983. "Debate on Learning Theory is Shifting," Science 222: 1219-22.
Frazer, J. 1955. The Golden Bough: A Study in Magic and Religion, 3rd ed. New York: St. Martin's
Press. Originally published in 1890.
Freeman, D. 1983. Mr~.rgaret Mead and Samoa: The Mrlldng and Unmaldng of an Anthropological
Myth. Cambridge, Mass.: Harvard University Press.
Fries, C. C. 1963. "The Bloomfield School." In C. Mohrmann, A. Sommerfelt, and J. What-
mough, eds., Trends in Europerm Rnd American Linguistics 1930-1960. Utrecht: Spectrum.
Fritsch, G.; and Hitzig, E. 1870. "Uber die elektrische Erregbarkeit des Grosshims," Arch. AnRt.
Physiol. u. Wisse. Med. 37.
Fryer, D. M.; and Marshall, J. C. 1979. "The Motives of Jacques de Vaucanson," Technology
and Culture 20: 257-69.
Gardin, J. C. 1965. "On a Possible Interpretation of Componential Analysis in Archaeology,"
American Anthropologist 67: 9-11.
Gardner, H. 1975. The Shattered Mind: The Person after Brain Dtlmage. New York: Alfred A. Knopf.
- - . 1978. Developmental Psychology. Boston: Uttle, Brown.
- - - . 1983. FrRmes of Mind: The Theory of Multiple Intelligences. New York: Basic Books.
Gazdar, G. 1981. "On syntactic categories," Philosophical TrRnsactions of the Royal Society of Lon-
don, Series B. 295: 267-83.
Gazzaniga, M. S. 1981. "Nobel Prize for Physiology or Medicine," Science 214: 517-18.
- - - . 1983. "Right Hemisphere Language Following Commisorotomy: A Twenty-Year
Perspective," American Psychologist 38: 525-37.
Geertz, C. 1973. The Interpretation of Cultures. New York: Basic Books.
- - - . 1983. Local Knowledge. New York: Basic Books.
Geschwind, N. 1965. "Disconnexion Syndromes in Animals and Man," Brain 88: 585-644.
- - - . 1974. Selected Papers on Language and the Brain. Dordrecht-Boston/Reidel.
Gibson,]. J, 1950. The Perception of the Vtsual World. Boston: Houghton-Mifflin.
- - - . 1966. The Senses Consirkred as Perceptual Systems. Boston: Houghton-Mifflin.
- - - . 1967. "New Reasons for Realism," Synthese, 17: 162-72.
- - - . 1979. The Ecological ApproRch to Vtsual Perception. Boston: Houghton-Mifflin.
Glass, A. L.; Holyoak, K. J.; and Santa,]. L. 1979. Cognition. Reading, Mass.: Addison-Wesley.
Goldman, P. S.; and Galkin, T. W. 1978. "Prenatal Removal of Frontal Association Cortex
397
References
in the Fetal Rhesus Monkey: Anatomical and Functional Consequences in Postnatal
Life," Brain Research 152: 451-85.
Goldman-Rakic, P. S.; lsseroff, A.; Schwartz, M. L.; and Bugbee, N. M. 1982. "Neurobiology
of Cognitive Development in Non-human Primates." Unpublished manuscript, Yale
School of Medicine, New Haven.
Goldstine, H. 1972. The Computer from PllSCal to oon Neumann. Princeton: Princeton University
Press.
Goodenough, W. H. 1951. Properly, Kin, and Community on Truk. New Haven: Yale University
Press.
· - - - . 1956. "Componential Analysis and the Study of Meaning," Langtmgt 32: 195-216.
- - - . 1964. "Language and Property in Truk: Some Methodological Considerations." In D.
H. Hymes, ed., Language in Culture and Society: A Reader in Linguistics and Anthropology. New
York: Harper & Row.
- - - . 1968. "Componential Analysis." In D. Sills, ed., International Encyclopedia of the Social
Sciences. New York: Macmillan.
Goodman, N. 1955. Fact, Fiction and Forecast. Cambridge, Mass.: Harvard University Press.
- - - . 1972. "The Revision of Philosophy." InN. Goodman, Problems and Projects. Indianapo-
lis: Bobbs-Merrill. Originally published in 1963.
Gross, C. G.; Rocha-Miranda, C. E.; and Bender, D. 1972. "Visual Properties of Neurons in
Inferotemporal Cortex of the Macaque," Journal of Neurophysiology 35: 96-111.
Grossberg, S. 1980. "How Does a Brain Build a Cognitive Code?," Psychological R1!11iew 87: 1-51.
Gruber, H. E.; and Vom!che, J. J., eds. 1977. The Essential Piaget. New York: Basic Books.
Hacking, I. 1980. "Is the End in Sight for Epistemology?," j(}Urnal of Philosophy 77: 579-88.
- - - . 1982. "Wittgenstein the Psychologist" [Review of Remarks on the Philosophy of Psy-
chology, G. E. M. Anscombe, ed.], New York Review of Books, 1 April.
Halacy, D. S., Jr. 1962. Computers the Machines We Think With. New York: Dell.
Hanfling, 0., ed. 1972. Fundamental Problems in Philosophy. Blackwell: Open University Press.
Harris, M. 1968. The Rise of Anthropological Theory: A Htstory of Theories of Culture. New York:
Thomas Y. Crowell.
Harris, R. 1981. The Language Myth. London: Duckworth.
Harris, W. H.; and Levy, J. S., eds. 1975. "Holography." The New Columbia Encyclopedia. New
York: Columbia University Press.
Harris, Z. S. 1952. "Discourse Analysis," Language 28: 18--30.
Harth, E. 1982. Windows on the Mind: Reflections on the Physical Basis of Consciousness. New York:
William Morrow.
Haugeland, J. 1981. "The Nature and Plausibility of Cognitivism." In J. Haugeland, ed., Mind
Design: Philosophy, Psychology, Artificial Intelligence. Montgomery, Vt.: Bradford Books.
Havelock, E. A. 1982. Preface to Plato. Cambridge, Mass.: Belknap Press/Harvard University
Press.
Hayes, J. R. 1978. Cognitive Psychology: Thinking and Creating. Homewood, Ill.: Dorsey Press.
Hayes, P. 1982. Personal communication. March 1982.
Hebb, D. 0. 1949. Organization of Behavior. New York: John Wiley.
Heider, E. R. 1972. "Universals in Color Naming and Memory," ](}U17Ull of frperimenflll Psy-
chology 93: 10-20.
---;and Olivier, D. C. 1972. "The Structure of the Color Space in Naming and Memory
for Two Languages," Cognitive Psychology 3: 337-54.
Heims, S. J. 1980. fohn von Neumann and Norbert Wiener. Cambridge, Mass.: MIT Press.
Helmholtz, H. von. 1962. Treatise on Physiological Optics. Vol. Ill J.P. C. Southall, trans. New York:
Dover.
Herrnstein, R. J.; and Boring, E. G., eds. 1965. A Source Book in the History ofPsychology. Cambridge,
Mass.: Harvard University Press.
Herskovitz, M. J. 1953. Franz Boas: The Science of Man in the Making. London: Charles Scribner.
Hilton, A. M. 1963. Logic, Computing Machines, and Automation. New York: World Publishing.
Hinton, G. E. 1981. "Implementing Semantic Networks in Parallel Hardware." In G. E. Hinton
and J, A. Anderson, eds., Parallel Models of Associative Memory. Hillsdale, N.J.: Lawrence
Erlbaum.
---;and Anderson, J. A. 1981. Parallel Models of Associative Memory. Hillsdale, N.J.: Lawrence
Erlbaum.
Hinton, G. E.; Sejnowski, T. J.; and Ackley, D. H. 1984. "Boltzmann Machines: Constraint
Satisfaction Networks That Learn." Computer Science Memo, Carnegie-Mellon University.
398
References
Hjelmslev, L. 1953. "Prologomena to a Theory of Language," IndiRnR Publimlions in Anthropology
Rnd Linguistics, no. 7.
Hochberg, J. 1978. Perception. Englewood Cliffs, N.J.: Prentice-Hall.
Hockett, C. 1958. A Course in Modem Linguistics. New York: Macmillan.
- - . 196&. Article on Leonard Bloomfield. In D. L. Sills, ed., lnlmzRtionRI Encyc/opediR of tht
SodRI Sdenm, pp. 95-98. New York: Macmillan.
- - . 1968b. Tht Shlte of the Art. The Hague: Mouton.
Hodgen, M. T. 1964. Early Anthropology in the Sixteenth Rnd Sromtemth Centuries. Philadelphia:
University of Pennsylvania Press.
Hodges, A. 1983. Aum Turing: Tht EnigmR. New York: Simon & Schuster.
Hofstadter, D. R. 1983. "Artificial Intelligence: Subcognition as Computation." In F. Machlup
and U. Mansfield, eds., The Study of ln/ormRfion: lnltrdisdplinRry MtsSRgtS. New York: John
Wiley.
--;and Dennett, D. 1981. The Mind's 1 New York: Basic Books.
Holton, G. 1984. "Do Scientists Need a Philosophy?," Tht Times Literary Supplnnml, 2. November
1984, pp. 2.31-34.
Holyoak, K. 1983. "Toward a Unitary Theory of Mind" [Review of J. R. Anderson, The
Architecture of Cognition], Science 2.2.2: 499-500.
Hooper, J. 1982.. Interview with Karl Pribram. Omni, October 1982, pp. 129-35, 169-76.
Horn, B. K. P. 1975. "Obtaining Shape from Shading Information." In P. Winston, ed., The
Psychology of Compultr V'tsion. New York: McGraw-Hill.
Hubel, D. H. 1978. "Vision and the Brain," Bu/lefin of the AmericRn Academy of Arts and
Sdenm 31: 17-28.
- - . 1979.' "The Brain," Scientific American 241: 44-53.
--;and Wiesel, T. N. 1959. "Receptive Fields of Single Neurones in the Cat's Striate
Cortex," fournRl of Physiology 148: 574-91.
Hubel, D. H.; and Wiesel, T. N. 1962. "Receptive Fields, Binocular Interaction and Functional
Architecture in the Cat's Visual Cortex," JournRI of Physiology 160: 10~4.
- - . 1979. "Brain Mechanisms of Vision," Scientific American 241: 150--62.
Hume, D. 1739. A Tmtise of HumRn Nature. London.
- - . 1955. An lnl{uiry Concerning Human Undershlnding. New York: Liberal Arts Press. Original
work published in 1748.
Humphrey, G. 1951. Thinking: An Introduction to ExpmmmtRl Psychology. New York: John Wiley.
Hunt, M. 1982. The Universe Within. New York: Simon & Schuster.
Hutchins, E. 1980. Culture Rnd Inference: A TrobriRnd CRSe Study. Cambridge, Mass.: Harvard
University Press.
Hymes, D. 1974. Studies in the History of Linguistics. Bloomington: Indiana University Press.
lnhelder, B.; and Piaget, J. 1958. Tht Grf!Wth of Logiml Thinking from Childhood to Adolescence, trans.
by A. Parsons and S. Milgram. New York: Basic Books.
- - . 1964. Tht Early Grf!Wth of Logic in the Child: ClllSSi/imtion Rnd SeriRtion. London: Routledge
& Kegan Paul.
Jackendoff, R. 1983. SemRnfics and Cognition. Cambridge, Mass.: MIT Press.
Jackson, J. Hughlings. 1932. Selected Writings. London: Hodder & Stoughton.
Jakobson, R. 1941. KindersprRChe, AphRSie und Rl/gemeine LRutgesetze. Uppsala: Universitets Aarssk-
rift. English ed.: Child LRngugge, AphRSiR Rnd PhonologicRl UnivtrsRls. The Hague: Mouton, 1968.
- - . 1963. Essais de linguislil{ue gblb-Rle. Paris: Editions de minuit.
- - ; and Halle, M. 1956. Furulamenhl/s of LRngu~~ge. The Hague: Mouton.
James, W. 1890. The Principles of Psychology. New York: Henry Holt.
Jefferson, G. 1949 "The Mind of Mechanical Man," Lister oration for 1949, British Mediml
fourn~~l 1: 1105-21.
Jeffress, L. A., ed. 1951. Cerebral Mechanisms in Behar:lior. Tht HIXon Symposium. New York: John
Wiley.
Johnson, M. K.; Bransford, J.D.; and Soloman, S. 1973. "Memory for Tacit Implications of
Sentences," ]ournRI of Ezperimenhll Psychology 98: 203-5.
Johnson-Laird, P. N. 1982. "Propositional Representations, Procedural Semantics, and Mental
Models." In J. Mehler, E.C.T. Walker, and M. Garrett, eds., Perspecfi!lts on MmtRl Represmhl·
lion: Experimmhll Rnd Theoretical Studies of Cognifi'Of Processes and CRpadties. Hillsdale, N.J.: Law-
rence Erlbaum.
- - . 1983. Menhll Models: TOWRrds a Cognifiue Sdmce of LRngu~~gt, Inference, Rnd Consdousness.
Cambridge, Mass.: Harvard University Press.
399
References
---;and Wason, P. C. 1970. "A Theoretical Analysis of Insight into a Reasoning Task,"
Cogniti'Of Psychology 1: 134-48.
Johnson-Laird, P. N.; and Wason, P. C., eds. 1977. Thinking: Readings in Cogniti'Of Science. Cam-
bridge: Cambridge: University Press.
Kahneman, D.; Slovic, P.; and Tversky, A., eds. 1982. judgement under Uncertainty: Heuristics and
Biases. New York: Cambridge University Press.
Kahneman, D.; and Tversky, A. 1982. "The Psychology of Preferences," Scientific American 246:
160-74.
- - - . 1984. "Choices, Values, and Frames," American Psychologist 39: 341-50.
Kandel, E. R. 1979. "Small Systems of Neurons," Scientific American 241: 66--84.
- - - . 1982. "Steps toward a Molecular Grammar for Learning: Explorations into the Nature
of Memory." Paper presented at the Bicentennial Symposium of the Harvard Medical
School, October 1982.
Kant, I. 1958. Critil{lit of Pure Reason, trans. by N. Kemp Smith. New York: Random House.
Original work published in 1781.
Kardiner, A.; and Preble, E. 1961. They Studied Man. Oeveland: World Publishing.
Katz, J.; and Fodor, J. 1963. "The Structure of a Semantic Theory," Language 39: 170-210.
Kay, P. 1966. "Comment on B. N. Colby, Ethnographic Setrulntics: A Preliminary Survey," Current
Anthropo{ggy 7: 3-32.
Keesing, R. M. 1976. Cultural Anthropology: A Contemporary Perspecti'Of. New York: Holt, Rinehart
& Winston.
Keil, F. C. 1979. Semantic and Conceptual Development: An Ontological Perspective. Cambridge, Mass.:
Harvard University Press.
Kenny, A. 1973. Witfgmsfein. Harmondsworth, Middlesex, England: Penguin Books.
Kessen, W. 1981. "Early Settlements in New Cognition," Cognition 10: 167-71.
Kimura, D. 1973. "The Asymmetry of the Human Brain," Scimtific American 228: 70-80.
Kinsbourne, M., ed. 1978. Assymefrical Function of the Brain. Cambridge: Cambridge University
Press.
Kitto, H. D. F. 1951. The Greeks. Middlesex, England: Penguin Books.
Kohler, W. 1925. The MenfRiify of Apes, trans. by E. Winter. New York: Humanities Press.
(Original work published in 1917.)
- - - . 1969. The Task of Gestalt Psychology. Princeton: Princeton University Press.
Konishi, M. 1969. "Experimental Studies in the Ontogeny of Avian Vocalizations." In R. A.
Hinde, ed., Bird Vocalization. Cambridge: Cambridge University Press.
Kosslyn, S.M. 1978. "Imagery and Internal Representation." In E. Rosch and B. B. Lloyd, eds.,
Cognition and Categorization. Hillsdale, N.J.: Lawrence Erlbaum.
- - - . 1980. Image and Mind. Cambridge, Mass.: Harvard University Press.
- - - . 1981. "The Medium and the Message in Mental Imagery: A Theory," Psychological
Review 88: 46-66.
- - - . 1983. Ghosts in the Mind's Machine: Creating and Using Images in the Brain. New York: W.
W. Norton.
- - - . 1984. "Imagery and Cerebral Lateralization: A Computational Theory." Unpublished
paper, Harvard University.
---;Ball, T. M.; and Reiser, B. J, 1978. "Visual Images Preserve Metric Spatial Information:
Evidence from Studies of Image Scanning," journal of Experimental Psychology: Hui1Uln Perception
and Performance 4: 47-60.
Kosslyn, S.M.; Pinker, S.; Smith, G. E.; and Schwartz, S. P. 1979. "On the Demystification
of Mental Imagery," The Behauioral and Brain Scienas 2: 535-81.
Kripke, S. 1972. "Naming and Necessity." In D. Davidson and G. Harman, eds., Stmantics of
Natural Language. Dordrecht/Boston: Reidell.
Kroeber, A. 1917. "The Super-organic," Americ11n Anthropologist 17: 163-213.
- - - . 1948. Anthropology. New York: Harcourt, Brace.
Kroeber, T. 1970. A/fred Kroeber: A Personal Configur11tion. Berkeley: University of California
Press.
Kuhn, T. 1970. The Structure of Scientific Revolution, 2nd ed. Chicago: University of Chicago Press.
Kuper, A. 1973. Anthropologists and Anthropology: The British School 1922-1972. New York: Pica
Press.
Kyburg, H. E. 1983. "Rational Belief," The Behavioral and Br11in Sciences 6: 231-73.
Labbe, R.; Firl, A.; Mufson, E.; and Stein, D. 1983. "Fetal Transplants: Reduction of Cognitive
Deficits in Rats with Frontal Cortex Lesions," Science 221: 470-72.
400
References
Lachman, R.; Lachman, J. L.; and Butterfield, E. C. 1979. Cognitive Psychology and Information
Processing: An Introduction. Hillsdale, N.J.: Lawrence Erlbaum.
Lakoff, G. 1969. "Empiricism without Facts," Foundations of Language 5: 118-27.
- - - . 1980. "Whatever Happened to Deep Structure?" The Behavioral and Brain Sciences 3:
22-23.
- - . 1982. "Categories and Cognitive Models." Working paper. Berkeley Cognitive Sci-
ence Report no. 2.
--;and Ross, J. R. 1976. "Is Deep Structure Necessary?" In J. McCawley, ed., Syntu: and
Semantics, vol. 7. New York: Academic Press. Originally written in 1967.
Lashley, K. S. 1929. Brain Mechanisms and Intelligence. Chicago: University of Chicago Press.
- - . 1950. "In Search of the Engram." In Symposia of the Scdefy for Experimental Biology 4:
454-82.
Lashley, K. S.; Chow, K. L.; and Semmes, J. 1951. "An Examination of the Electrical Field
Theory of Cerebral Integration," Psychological Review 58: 123-36.
Lave, J. 1977. "Cognitive Consequences of Traditional Apprenticeship Training in West
Africa," Anthropology and Education Qu11rferly 8: 177-80.
Leach, E. 1961. "Levi-Strauss in the Garden of Eden: An Examination of Some Recent
Developments in the Analysis of Myth," TransiiCtions of the New York Academy of Sciences
D, 23: 386-96.
- - - . 1974. Uvi-SirRU$5, rev. ed. London: Fontana.
Leary, D. E. 1978. "The Philosophical Development of the Conception of Psychology in
Germany, 1780-1850," JournAl of the Htsfory of the Behavioral Sciences 14: 113-21.
Lees, R. B. 1957. "Review of Noam Chomsky," Syntactic Structures, LRnguage 33: 375-408.
Lenat, D. B. 1976. "AM: An Artificial Intelligence Approach to Discovery in Mathematics as
Heuristic Search." Report STAN-CS-76-570. Stanford University, Computer Science
Department.
- - - . 1984. "Computer Software for Intelligent Systems," Scientific American 251: 204-13.
Lettvin, J. Y. 1981. "Filling Out the Forms: An Appreciation of Hubel and Wiesel," Sci-
ence 214: 518-20.
- - ; Maturana, H. R.; McCulloch, W. S.; and Pitts, W. H. 1959. "What the Frog's Eye Tells
the Frog's Brain," Proceedinss of the IRE 47: 1940-51.
Levi-Strauss, C. 1963. Structural Anthropo/ggy, trans. by C. Jacobson and B. Grundfest Schoepf.
New York: Basic Books. Original work published 1958.
- - - . 1964. Tristes fropiques, trans. by J. Russell. New York: Atheneum. Original work
published in 1955.
- - . 1969. The Raw 11nd the Cooked, trans. by J. and D. Weightman. New York: Harper & Row.
Ughthill, J. 1972. "A Report on Artificial Intelligence." Unpublished manuscript, Science
Research Council.
- - ; Sutherland, N. S.; Needham, R. M.; Longuet-Higgins, H. C.; and Michie, D. 1973.
Artificial Intelligence: A Paper Symposium. London: Science Research Council.
Locke, J. 1975 An Essay Concerning Human Understanding. P. H. Nidditch, ed. Oxford: Clarendon
Press. Original work published 1690.
Longuet-Higgins, H. C. 1979. "The Perception of Music," Proceedings of the Royal Society of London
series B 205, pp. 307-22.
- - - . 1981. "Artificial Intelligence-A New Theoretical Psychology?," Cognition 10: 197-
200.
- - - . 1982. "A Theory of Vision" [review of David Marr's Vision), Science 218: 991-92.
- - - ; Lyons, J.; and Broadbent, D. 1981. The Psychological Mechanisms of Language. A joint
symposium of the Royal Society and the British Academy, London.
Lorenz, K. 1935. "Der Kumpan in der Umwelt des Vogels," Journal of Ornithology 87: 137-413.
Lounsbury, F. G. 1956. "A Semantic Analysis of the Pawnee Kinship Usage," Language 32:
158-94.
- - - . 1963. "Linguistics and Psychology." In S. Koch, ed., Psychology: A Study of a Science. New
York: McGraw-Hill.
- - . 1968. "One Hundred Years of Anthropological Unguistics." In J. 0. Brew, ed., One
Hundred Years of Anthropology. Cambridge, Mass.: Harvard University Press.
Lovejoy, A. 0. 1964. The Great Chllin of Being. Cambridge, Mass: Harvard University Press.
Lowry, R. 1971. The Evolution of PsychologicRl Theory 1650 to Present. Chicago: Aldine-Ather-
ton.
401
References
Luce, R. D.; Bush, R. R.; and Galanter, N. F., eds. 1963. Handbook of Mathematical Psychol-
ogy. New York: John Wiley.
Luchins, A. S. 1942. "Mechanization in Problem Solving," Psychological Monographs 54(6), whole
no. 248.
Luria, A. R. 1966. Higher Cortical Functions in Man. New York: Basic Books.
Lyons, J. 1968. lnlroduction to Theoretical Linguistics. Cambridge: Cambridge University Press.
- - - . 1970. Noam Chomsky. New York: Viking Press.
McArthur, D. J. 1982. "Computer Vision and Perceptual Psychology," Psychowgical Bulletin 92:
283-309.
McCarthy, J. 1984. Interview, Stanford University, February 1984.
- - ; Abrahams, P. W.; Edwards, D. J.; Hart, T. D.; and Levin, M. I. (1962). Lisp I. 5
Programmn's Manual. Cambridge, Mass.: MIT Press.
---;and Hayes, P. J. 1969. "Some Philosophical Problems from the Standpoint of Artificial
Intelligence." In B. Meltzer and D. Michie, eds., Machine Intelligence 4. Edinburgh: Edin-
burgh University Press.
McCorduck, P. 1979. Machines Who Thinlr. San Francisco: W. H. Freeman.
McCulloch, W.; and Pitts, W. 1943. "A Logical Calculus of the Ideas Immanent in Nervous
Activity," Bulletin of Mathematical Biophysics 5: 115-33.
McDougall, W. 1961. Body and Mind. Boston: Beacon Press. Original publication in 1911.
Malinowski, B. 1961. Argonauts of the Western-Pacific: An Account of Natiue Enterprise and Adventure
in the Archipelagoes of Meltmesian N!!W Guinea. New York: E. P. Dutton. Original work
published in 1922.
- - - . 1968. Culture. In D. L. Sills, ed., International Encyclopedia of the Social Sciences. New York:
Macmillan. Original work published in 1935.
Mandler, G. 1981. "What Is Cognitive Psychology? What Isn't?" Address to the APA Divi-
sion of Philosophical Psychology, Los Angeles.
Mandler, J. M.; and Mandler, G. 1964. Thin/ring: From Association to GesfRlt. New York: John
Wiley.
Marcus, M. 1979. "A Theory of Syntactic Recognition for Natural Language." In P. H.
Winston and R. H. Brown, eds., Artificial Intelligence: An MIT Perspectiue, vol. 1, pp. 191-229.
Cambridge, Mass.: MIT Press.
- - - . 1980. A Theory of Syntactic Recognition for Natural Language. Cambridge, Mass.: MIT Press.
Marie, P. 1906. "Revision de la question de l'aphasie," Semaine MMicale 21: 241-47.
Marler, P. 1982. "Avian and Primate Communication: The Problem of Natural Categories,"
Neuroscience and Biobehavioral Revii!WS 6: 87-94.
- - - ; and Peters, S. 1977. "Selective Vocal Learning in a Sparrow," Science 198: 519-21.
Marr, D. 1969. "A Theory of Cerebellar Cortex," journal of Physiology 202: 437-470 (London).
- - - . 1979. "Representing and Computing Visual Information." In P. H. Winston and R.
H. Brown, eds., Artificial Intelligence: An MIT Perspectiue, vol. II: Understanding V"tsion, Manipula-
tion, Computer Design, Symbol Manipulation. Cambridge, Mass.: MIT Press.
- - - . 1982. V"tsion: A Computational Investigation info the Human Representation and Processing of Visual
Information. San Francisco: W. H. Freeman.
---;and Nishihara, K. 1978. "Representation and Recognition of the Spatial Organization
of Three-Dimensional Shapes," Proceedings of the Royal Society London, B, 200, pp. 269-94.
Marshall, J. C. 1980. "Artificial Intelligence-The Real Thing?," The Behavioral and Brain
Sciences 3: 435-37.
- - - . 1981. "Cognition at the Crossroads" [review of M. Piattelli-Palmarini, ed., Language
and Learning: The Debate between Jean Piaget and Noam Chomsky], Nature 289: 613-14.
Matthews, J. R. 1982. "Who's Afraid of Immanuel Kant?" [review of N. Block, ed., Readings
in Philosophy of Psychology, vols. I, II], Contemporary Psychology 27: 433-34.
Mehler, J.; Morton, J.; and Jusczyk, P. 1984. "On Reducing Language to Biology," Cognitiue
Neuropsychology 1: 83-116.
Mehler, J.; Walker, E. C. T.; and Garrett, M., eds. 1982. Perspectives on Mental Representation:
Experimental and Theoretical Studies of Cognitiue Processes and Capacities. Hillsdale, N.J.: Lawrence
Erlbaum.
Mervis, C. B.; and Rosch, E. 1981. "Categorization of Natural Objects," Annual Review of
Miller, G. A. 1956. "The Magical Number Seven, Plus or Minus Two: Some Limits on Our
Capacity for Processing Information," Psychological Revii!W 63: 81-97.
- - - . 1962. "Some Psychological Studies of Grammar," American Psychologist 17: 748-62.
402
References
- - - . 1979. "A Very Personal History." Talk to Cognitive Science Workshop, Massachu-
setts Institute of Technology, Cambridge, Mass. 1 June 1979.
- - - . 1982. Interview, Princeton University, January 1982.
- - - ; Galanter, E.; and Pribram, K. 1960. Plans and the Structure of Behavior. New York: Holt,
Rinehart & Winston.
Miller, J. 1983. Slates of Mind. New York: Pantheon Books.
Milner, B. 1967. "Brain Mechanisms Suggested by Studies of Temporal Lobes." In F. L.
Darley, ed., Brain Mechanisms Underlying Speech and Language. New York: Grune & Strat-
ton.
Milner, P. M. 1970. Physiological Psychology. New York: Holt, Rinehart & Winston.
Minsky, M. 1963. "Steps toward Artificial Intelligence." In E. A. Feigenbaum and J. Feldman,
eds., Computers and Thought. New York: McGraw-Hill. Originally circulated in 1957.
- - - , ed. 1968. Semantic Information Processing. Cambridge, Mass: MIT Press.
- - - . 1975. "A Framework for Representing Knowledge." In P. H. Winston, ed., The
Psychology of Computer Vision. New York: McGraw-Hill.
- - . 1979. "The Society Theory." In P. H. Winston and R. H. Brown, eds., Artificial
lnfelligence: An MIT Perspective, vol. 1. (pp. 423-50). Cambridge, Mass.: MIT Press.
- - - . 1982. "Learning Meaning." Unpublished manuscript, Massachusetts Institute of
Technology.
---;and Papert, S. 1968. Percepfrons. Cambridge, Mass.: MIT Press.
Mishkin, M. 1967. "Visual Mechanisms beyond the Striate Cortex." In R. W. Russell, ed.,
Frontiers in Physiological Psychology. New York: Academic Press.
Moates, D. R.; and Schumacher, G. M. 1980. An Introduction to Cognitive Psychology. Belmont,
Calif.: Wadsworth.
Mohrmann, C.; et al. 1963. Trmds in Modern Linguistics. Utrecht: Spectrum.
Moray, N. 1969. Lisfming and Attention. Baltimore, Md.: Penguin.
Morris, C. 1971. Writings on the Gmeral Theory of Signs. The Hague: Mouton.
Morton, J. 1981. "Will Cognition Survive?" Cognition 10: 227-34.
Mountcastle, V. B. 1978. "An Organizing Principle for Cerebral Function: The Unit Module
and the Distributed System." In G. M. Edelman and V. B. Mountcastle, eds., The Mindful
Brain. Cambridge, Mass.; MIT Press.
Murray, H. A. 1945. "Assessment of the Whole Person," Proceedings Meeting Military Psychologists
and Psychiatrists. University of Maryland Press.
Murray, S. 0. 1982. "The Dissolution of 'Classical Ethnoscience,' "fourmzl of fhe History of fhe
Behavioral Sciences 18: 63-175.
Neisser, U. 1967. Cognitive Psychology. New York: Appleton-Century-Crofts.
- - - . 1976. Cognition and Reality. San Francisco: W. H. Freeman.
- - - . 1980. "Comments on Gestalt Psychology." Paper presented at APA Symposium,
September 1980.
- - - . 1984. "Toward an Ecologically Oriented Cognitive Science." Emory Cognition Project,
Report 1, Atlanta, Georgia, January 1984.
Newell, A. 1983. "Intellectual Issues in the History of Artificial Intelligence." In F. Machlup
and U. Mansfield, eds., The Study of Information: Interdisciplinary Messages. New York: John
Wiley.
- - - ; Shaw, J. C.; and Simon, H. A. 1964. "Elements of a Theory of Human Problem-
solving." In R. J. C. Harper, C. C. Anderson, C. M. Christensen, and S.M. Hunka, eds.,
The Cognitive Processes: Readings. Englewood Cliffs, N.J.: Prentice-Hall. Reprinted from Psy-
chological Review 65 (1958): 151-66.
Newell, A.; and Simon, H. A. 1972. Human Problem Solving. Englewood Cliffs, N.J.: Prentice-
Hall.
Newmeyer, F. 1980. Linguistic Theory in America: The First Quarter Century of Transformational Genera-
five Grammar. New York: Academic Press.
Nilsson, N. J. 1980. Principles of Artificial Intelligence. Palo Alto, Calif.: Tioga.
Norman, D. 1980. "Twelve Issues for Cognitive Science,'' Cognitive Scimce 4: 1-32.
Nottebohm, F. 1970. "Ontogeny of Bird Song," Science 167: 950-56.
- - - . 1980. "Brain Pathways for Vocal Learning in Birds: A Review of the First 10 Years,''
Progress in Psychobiological and Physiological Psychology 9: 85-124.
Osherson, D. N.; and Smith, E. E. 1981. "On the Adequacy of Prototype Theory as a Theory
of Concepts," Cognition 9: 35-58.
Paivio, A. 1971. Imagery and Verbal Processes. New York: Holt, Rinehart & Winston.
403
References
Peters, S.; and Ritchie, R. 1973. "On the Generative Power of Transformational Grammars,"
Information Sciences 6: 49-83.
Piaget, J. 1970. "Piaget's Theory." In P. Mussen, ed., Carmichael's Manual of Child Psychology
l New York: John Wiley.
Piattelli-Palmarini, M., ed. 1980. Language and laming. Cambridge, Mass.: Harvard University
Press.
- - . 1983. "The Parliament of Mind" [review of J. A. Fodor, The Modularity of Mind], The
Times Lilerflry Suppltmmf, 2 December 1983, p. 1357.
Pinker, S. 1984. Language Uflmability and languflgt DnM/opment. Cambridge, Mass.: Harvard Uni-
versity Press.
Plato. 1956fl. Mmo. In E. H. Warmington and P. 0. Rouse, eds., Great Dialogues of Plato, trans.
by W.H.D. Rouse. New York: New American Library.
- - . 1956b. Thfletdus. In I. Edman, ed., The Philosophy of Plato, trans. by B. Jowett. New York:
Random House.
Popper, K. 1974. Unendd Quest: An Intellectual AutobWgmphy. Glasgow, Scotland: William Col-
lins.
Posner, M. 1969. "Abstraction and the Process of Recognition." In G. H. Bower and J. T.
Spencer, eds., The Psychology of laming and Moti'(}(lfion, vol. 3. New York: Academic Press.
- - - ; and Mitchell, R. F. 1967. "Chronometric Analysis of Classification," Psychological
Rtview 74: 392-409.
Posner, M.; and Shulman, G. L. 1979. "Cognitive Science." In E. Hearst, ed., The First Century
of Experimenftll Psychology, Hillsdale, N.J.: Lawrence Erlbaum.
Pribram, K. H. 1971. LangullgtS of the BrRin: Experimenftll Paradoxes Rnd Principles in Neuropsy-
chology. Englewood Cliffs, N.J.: Prentice-Hall.
- - - . 1980. "Mind, Brain, and Consciousness: The Organization of Competence and Con-
duct." In R. J. Davidson and M. Davidson, eds., The Psychology of Consciousness. New York:
Plenum Press.
- - - . 1982. "Localization and Distribution of Function in the Brain." In J. Orbach, ed.,
Neuropsychology after Lashley. Hillsdale, N.J.: Lawrence Erlbaum.
Putnam, H. 1960. "Minds and Machines." InS. Hook, ed., Dimensions of Mind. New York: New
York University Press.
- - - . 1973. "Philosophy and Our Mental Life." Paper presented at Foerster Symposium on
Computers and the Mind at University of California at Berkeley. In H. Putnam, Mind,
Language and Reality: Philosophical Papers, vol. 2. Cambridge: Cambridge University Press,
1975.
- - . 1975a. Mind, Language 11nd Reality: Philosophical PRpers, vol. 2. Cambridge: Cambridge
University Press.
- - . 1975b. "The Meaning of 'Meaning."' (Reprinted in Putnam, 1975, pp. 215-71.)
- - - . 1981. "Why Reason Can't Be Naturalized." Howison Lecture, University of Califor-
nia at Berkeley, 30 April1981. Reprinted in Putnam, 1983.
- - - . 1983. Realism and Reason: Philosophical P11pers, vol. 3. Cambridge: Cambridge University
Press.
- - - . 1984. "After Ayer, after Empiricism," PRrtisan Review 2: 265-75.
Pylyshyn, Z. W. 1973. "What the Mind's Eye Tells the Mind's Brain: A Critique of Mental
Imagery," Psychological Bulletin 8: 1-14.
- - - . 1979. "Imagery Theory: Not Mysterious-Just Wrong," The BehlluiOrRl 11nd Brain
Sciences 2: 561-62.
- - - . 1980a. "The 'Causal Power' of Machines" [commentary on J. Searle, "Minds, Brains
and Programs'1, The &h11vioraland Br11in Sciences 3: 442-44.
- - - . 1980b. "Cognitive Representation and the Process-architecture Distinction," The
Beh11uioral Rnd BrRin Sciences 3: 154-69.
- - - . 1981. "The Imagery Debate: Analogue Media versus Tacit Knowledge," Psychological
Review 88: 16-45.
- - - . 1984. Compuftltion and Cognition: ToWRrd a Foundation for Cognitiut Science. Cambridge, Mass.:
MIT Press.
Quillian, M. R. 1968. "Semantic Memory." In M. Minsky, ed., Semantic Information Process-
ing. Cambridge, Mass.: MIT Press.
Quine, W. V. 0. 1953. From a Logical Point of View. Cambridge, Mass.: Harvard University Press.
Radcliffe-Brown, A. R. 1940. "On Social Structure," founu~l of the Royal Anthropologic11l Insti-
tute 70.
404
References
- - - . 1952. Structure Rnd Function in PrimifiTJe Society. Glencoe, n.: Free Press.
Rapaport, D. 1951. OrgtmizRfion Rnd PRthology of Tlwught. New York: Columbia University Press.
Reddy, D. R.; Erman, L. D.; Fennell, R. D.; and Neely, R. B. 1973. "The HEARSAY Speech
Understanding System." In Proceedings, Third InternRfion~tlfoint Conference on ArfijiciRl Intelli-
gence 3: 185-93 (Stanford).
Rey, G. 1981. "What Are Mental Images?" InN. Block, ed., lm11gery. Cambridge, Mass.: MIT
Press.
Rieber, R. W. 1983. DiRlogues on the Psychology of LRngullge Rnd Thought: ConversRtions with NORm
Chomsky, ChRrles Osgood, /l!lln PiRget, Ulric Ntisser, Rnd MRrcel Kinsbourne. New York: Plenum
Press.
Rivers, W. H. R. 1900. "A Genealogical Method of Collecting Social and Vital Statistics,"
fournRl of the AnthropologicRl lnstilute of Grl!llt BritRin Rnd Ireumd 30: 74-82.
Roberts, L. G. 1965. "Machine Perception of Three Dimensional Solids." InT. T. Tippett et
al., OpticRl Rnd Electro-opficRl lnformRtion Processing, Cambridge, Mass.: MIT Press.
Robins, R. H. 1967. A Short History of Linguistics. Bloomington: Indiana University Press.
Robinson, D. M. 1976. An lntelleciuRl History of Psychology. New York: Macmillan.
Romney, A. K.; and D' Andrade, R. G. 1964. "Cognitive Aspects of English Kin Terms,"
AmericRn Anthropologist 66: 146-70.
Rorty, R. 1979. Philosophy Rnd the Mirror of NR!ure. Princeton, N.J.: Princeton University Press.
- - - . 1982R. Personal communication. Princeton University, January 1982.
- - - . 1982b. "Mind as Ineffable." Unpublished manuscript, Princeton University.
Rosch, E. 1973R. "Natural Categories," CognitiTJe Psychology 4: 328-50.
- - - . 1973b. "On the Intemal Structure of Perceptual and Semantic Categories." InT. E.
Moore, ed., CognitirJe Development Rnd the A(fuisifion of LRngullgt. New York: Academic Press.
- - - . 1977. "Human Categorization." In N. Warren, ed., Aduances in Cross-culiuml Psychol-
ogy, vol. I. London: Academic Press.
- - - . 1978. "Principles of Categorization." In E. Rosch and B. B. Lloyd, eds., Cognition Rnd
CRiegorizRtion. Hillsdale, N.J.: Lawrence Erlbaum.
--;and Lloyd, B. B. 1978. Cognition Rnd GltegorizRtion. Hillsdale, N.J.: Lawrence Erlbaum.
Rose, S. 1973. The Conscious Bmin. New York: Alfred A. Knopf.
Rosenblueth, A.; Wiener, N.; and Bigelow, J. 1943. "Behavior, Teleology, and Purpose,"
Philosophy of Science 10: 18-24.
Rosenfield, I. 1984. "Seeing through the Brain," New York ReuinD of Books, 11 October 1984, pp.
53--56.
Royce, J. R.; and Rozeboom, W. W., eds. 1972. The Psychology of Knowing. New York: Gordon
& Breach.
Rumelhart, D. E. 1975. "Notes on a Schema for Stories." In D. G. Bobrow and A. M. Collins,
eds., RepresentRtion Rnd UnderstRnding. New York: Academic Press.
- - - . 1982. "Comments on Cognitive Science." Unpublished manuscript, University of
California at San Diego.
- - - ; and McClelland, J. L. 1982. "An Interactive Activation Model of Context Effects in
Letter Perception: part 2. The Contextual Enhancement Effect and Some Tests and
Extensions of the Model," PsychologicRl R1!11iew 89: 60-94.
Russell, B. 1961. History of Western Philosophy. London: George Allen & Unwin.
Ryle, G. 1949. The Concept of Mind. London: Hutchinson.
Sahlins, M. 1976. "Colors and Cultures," SmriotiCR 16: 1-22.
Sapir, E. 1921. lRngullgt. New York: Harcourt, Brace, & World.
- - - . 1929. "The Status of Linguistics as a Science." In Selected Writings of EdWRrd SRpir.
Berkeley: University of California Press.
Saussure, F. de. 1959. Course in Geneml Linguistics, ed. by C. Bailey and A. Sechehaye, in
collaboration with A. Riedlinger; trans. by W. Baskin. New York: Philosophical Library.
Originally published in 1915.
Schank, R. C. 1972. "Conceptual Dependency: A Theory of Natural Language Understand-
ing," CognitiTJe Psychology 3: 552-631.
- - - . 1980. "An Artificial Intelligence Perspective on Chomsky's View of Language," The
BehR~iOrRl Rnd Bmin Scimces 3: 35-37.
---;and Abelson, R. 1977. Scripts, PIRns, GORis, Rnd Undershlnding. Hillsdale, N.J.: Lawrence
Erlbaum.
Schwartz, R. 1981. "Imagery-There's More to It Than Meets the Eye." InN. Block, ed.,
f=gery. Cambridge, Mass.: MIT Press.
405
References
Schwartz, S. P. 1979. "Natural Kind Terms," Cognitiqn 7: 301-15.
Scribner, S. 1984. "Studying Working Intelligence." In B. Rogoff and J. Lave, eds., Ewryday
Cognitiqn. Cambridge, Mass.: Harvard University Press.
--;and Cole, M. 1981. The Psychology of Literacy. Cambridge, Mass.: Harvard University
Press.
Searle, J. 1980. "Minds, Brains and Programs," The Beha~ioral and Brain Sciences 3: 417-57.
- - - . 1983a. lnlentionalily: An Essay in the Philosophy of Mind. Cambridge: Cambridge University
Press.
- - - . 1983b. "Minds and Brains without Programs." Unpublished essay.
Segall, M. H.; Campbell, D. T.; and Herskovits, M. J. 1966. The Influence of Culture on VJSual
Perceptillrl. Indianapolis: Bobbs-Merrill.
Shannon, C. E. 1938. "A Symbolic Analysis of Relay and Switching Circuits." Master's thesis,
Massachusetts Institute of Technology; published in Trans11ctiqns of lhe American lnstilufe of
Electrical Engineers 57: 1-11.
--;and McCarthy,}., eds. 1956. AutomafR Studies; Annals of Mathematical Studies # 34. Prince-
ton: Princeton University Press.
Shanon, B. 1985. "The Role of Representation in Cognition." In J. Bishop, J. Lockhead, D.N.
Perkins, eds., Thinking. Hillsdale, N.J.: Lawrence Erlbaum.
Shebar, W. 1979. "Mental Imagery: A Critique of Cognitive Psychology." Paper submitted
for A.B. with Honors in Philosophy, Harvard University, Cambridge, Mass.
Shepard, R.N. 1981. "Psychophysical Complementarity." In M. Kubovy and J. R. Pomerantz,
eds., Perceptual Organizatillrl. Hillsdale, N.J.: Lawrence Erlbaum.
- - . 1982. "Perceptual and Analogical Bases of Cognition." In J. Mehler, E. C. T. Walker,
and M. Garrett, eds., Perspecti~ on MmfRl RepresenfRtiqn: EzperimenfRI11nd Theoretical Studies of
Cogniti~ Processes and Cap~~cities. Hillsdale, N.J.: Lawrence Erlbaum.
- - - ; and Chipman, S. 1970. "Second-order Isomorphism of Internal Representations:
Shapes of States," Cogniti~ Psychology 1: 1-17.
Shepard, R. N.; and Metzler, J. 1971. "Mental Rotation of Three-dimensional Objects,"
Science 171: 701-3.
Shiffrin, R. M.; and Schneider, W. 1977. "Controlled and Automatic Human Information
Processing: II. Perceptual Learning, Automatic Attending, and a General Theory," Psycho-
logiclll Review 84: 127-90.
Silverstein, M. 1978. "The Three Faces of Function: Preliminaries to a Psychology of Lan-
guage." In M. Hickmann, ed., Proceedings of a Working Conference on the Social Foundations of
Lllnguagt 11nd Thought. Chicago: Center for Psychosocial Studies.
Simon, H. A. 1969. Sciences of the Ar/ifici11l. Cambridge, Mass.: MIT Press.
- - - . 1982. Interview, Carnegie-Mellon University, February 1982.
---;and Newell, A. 1958. "Simulation of Cognitive Processes: A Report on the Summer
Research Training Institute, 1958." In Social Science Research Council Items 12: 37-40.
Skinner, B. F. 1957. Verbal Behllvior. New York: Appleton-Century-Crofts.
Sloan Foundation. 1976. "Proposed Particular Program in Cognitive Sciences." New York.
- - - . 1978. "Cognitive Science, 1978." Report of the State of the Art Committee. New
York.
Sloman, A. 1978. The Computer Rroolution in Philosophy: Philosophy, Science and Models of Mind.
Hassocks, Sussex: Harvester Press.
Smith, E. E.; and Medin, D. L. 1981. Categories and Concepts. Cambridge, Mass.: Harvard Univer-
sity Press.
Smith, N.; and Wilson, D. 1979. Modtrn Linguistics: The Results of Chomsky sRroolutillrl. Middlesex:
Penguin Books.
Sperber, D. 1968. Que'esl-ce que le structur11lismel 3. Lt Strucluralismt en anthropologie. Paris: Seuil.
- - - . 1982. Lt sawir des anlhropologues. Paris: Hermann.
Sperling, G. 1960. "The Information Available in Brief Visual Presentations," Psychological
Monographs 24: whole no. 11.
Sperry, R. W. 1951. "Mechanisms of Neural Maturation." InS. S. Stevens, ed., Handbook of
Ezperimenl11l Psychology. New York: John Wiley.
- - - . 1974. "Lateral Specialization in the Surgically Separated Hemispheres." In F. Schmitt
and F. G. Worden, eds., The Neurosciences: Third Study Program. Cambridge, Mass.: MIT Press.
- - - ; and Miner, N. 1955. "Pattern Perception Following Insertion of Mica Plates into
Visual Cortex," founu~l of Comparatiw 11nd Physiological Psychowgy 48: 463-69.
Sternberg, S. 1966. "High-Speed Scanning in Human Memory," Science 153: 652-54.
406
References
- - - . 1969. "Memory-Scanning: Mental Processes Revealed by Reaction-Time Experi-
ments," American Sdenfist 57: 421-57.
Stich, S. P. 1982. "The compleat cognitivist" [review of J. A. Fodor, Rqmsenftltions: Philosophical
Essays on the Foundations of Cognitiue Science], Confempora'71 Psychologist 27: 419-21.
- - . 1983. From Folk Psychology irJ Cognitiue Sdmce. Cambridge, Mass.: MIT Press (Bradford
Books).
- - - . 1984. "Thinking as per program" [review of P. N. Johnson-Laird, Mental Models:
Towrmis a Cognitive Sdence of Language, Inference, and ConsciOIISness ], Times Lifert~ry Supplement, 4
February 1984, p. 189.
Stocking, G. W. 1982. Rau, Culture, and Eoolufion: Essays in the Hislrlry of Anthropology. Chicago:
University of Chicago Press.
Stouffer, S. A.; et al. 1949. The American Soldier. Princeton, N.J.: Princeton University Press.
Sussman, G. J. 1975. A Computer Model of Skill Acl{uisifion. New York: American Elsevier.
Sutherland, S. 1982. "The Vision of David Marr" [review of D. Marr, VISion], Nature 298:
691-92.
Thomason, R., ed. 1974. Formal Philosophy: Selected Papers of Richard Montague. New Haven: Yale
University Press.
Thomson, R. 1968. The Pelican Htstory of Psychology. Middlesex, England: Penguin.
Tinbergen, N. 1951. The Study of lnslinct, London: Oxford University Press.
Tolman, E. C. 1932. Purposiue Behauior in Animals and Men. New York: Century.
Treisman, A.M. 1960. "Contextual Cues in Selective Listening," Quarterly Journal of Ezperimenflll
- - - . 1964. "Selective Attention in Man," British Medical Bulletin 20: 12-16.
Turing, A.M. 1936. "On Computable Numbers, with an Application to the Entscheidungs-
Problem," Proceedings of the London Mathematical Somty, Series 2, 42: 230-65.
- - - . 1963. "Computing Machinery and Intelligence." In E. A. Feigenbaum and J. Feldman,
eds., Computers and Thought. New York: McGraw-Hill. Original work published 1950.
Turvey, M. T.; Shaw, R. E.; Reed, E. S.; and Mace, W. M. 1981. "Ecological Laws of Perceiving
and Acting: In Reply to Fodor and Pylyshyn (1981)," Cognition 9: 237-304.
Tversky, A.; and Kahneman, D. 1983. "Extensional vs. Intuitive Reasoning: The Conjunction
Fallacy in Probability Judgment," Psychological Review 90: 293-315.
Tyler, S. A., ed. 1969. Cognitiue Anthropology. New York: Holt, Rinehart, & Winston.
- - - . 1978. T1te Said and the Unsaid: Mind, Meaning, and Culture. New York: Academic Press.
Tylor, E. 1871. Primitive Culture. London: John Murray.
Ullman, S. 1980. "Against Direct Perception," T1te Behauioral and Brain Sdmces 3: 373-416.
Vachek, J. 1966. The Linguistic School of Prague. Bloomington: Indiana University Press.
Van Lehn, K.; Seely Brown, J.; and Greeno, J. 1982. Competitive Argumentation in Compuflllionttl
Theories of Cognition (Interim report). Cognitive and Instructional Science Series. Xerox
Pare, Palo Alto.
Von Neumann, ]. 1958. The Computer and the Brain. New Haven, Conn.: Yale University
Press.
Waldrop, M. 1984a. "Artificial Intelligence (1): Into the World," Science 223: 802-5.
- - - . 1984b. "The Intelligence of Organizations," Science 225: 1136-37.
- - . 1984c. "Artificial Intelligence in Parallel," Science 225: 608-10.
Wallace, A. F. C. 1968. "Cognitive Theory." InD. L. Sills, ed., lnfernllfional Encyclopedia of the Social
Sciences. New York: Macmillan.
--;and Atkins, J. 1960. "The Meaning of Kinship Terms," American Anthropologist 62:
58-80.
Waltz, D. 1975. "Understanding Line Drawings of Scenes with Shadows." In P. H. Winston,
ed., The Psychology of Computer VISion. New York: McGraw-Hill.
Wason, P. C. 1966. "Reasoning." In B. Foss, ed., New Horizons in Psychology, vol. 1. Harmonds-
worth, Middlesex: Penguin.
- - ; and Johnson-Laird, P. N. 1972. The Psychology ofReasoning: Structure in Content. Cambridge,
Mass.: Harvard University Press.
Watson, J. B.; and Rayner, R. 1920. "Conditioned Emotional Reactions," Journal of Experimental
Psychology 3: 1-14.
Watson, R.I., ed. 1979. Basic Writings in the Hisirlry of Psychology. New York: Oxford University
Press.
Weimer, W. B. 1973. "Psycholinguistics and Plato's Paradoxes of the Meno," American
Psychologist, January, pp. 15-33.
407
References
Weiss, P. 1952.. "Central versus Peripheral Factors in the Development of Coordination,"
Research Publications of the Association for Research in Nm10us and Mental DistaSe 30: 3-2.3.
Weizenbaum, J. 1966. "ELIZA-A Computer Program for the Study of Natural Language
Communication between Man and Machine," Communications of the Association for Computing
Machinny 9: 36-45.
- - - . 1976. Computer Power and Human Rtason. San Francisco: W. H. Freeman.
Wernicke, C. 1874. Der aphasische sympfomen!ompla. Breslau: Cohn & Weigart.
Werthheimer, M. 1912.. "Experimentelle Studien iiber das Sehen von Bewegungen," Utshrift
for Psychologie 61: 161-2.65.
- - - . 1945. Productive ThinJ:ing. New York: Harper & Brothers.
Wexler, K. 1978. "A review of John Anderson, Liznguage, MetnllrY and Thought," Cognition 6:
32.7-51.
- - - . 1982.. "A Principle Theory for Language Acquisition." In L. Gleitman and H. E.
Wanner, eds., Liznguagt AclfUisition: The Stale of the Art. New York: Cambridge University
. Press.
--;and Culicover, H. 1980. Formal Principles of Language Acquisitwns. Cambridge, Mass.: MIT
Press.
White, L.A. 1963. "The Ethnography and Ethnology of Franz Boas," Memorial Museum Bulletin,
no. 6, Austin, TX.
Whitehead, A. N. 192.5. Science and the Modern World. New York: Macmillan.
- - - ; and Russell, B. 1910-13. Principia Mathnnatica. Cambridge: Cambridge University
Press.
WhorE, B. 1956. Liznguage, Thought and Reality, ed. by}. Carroll. Cambridge, Mass.: MIT Press.
Wiener, N. 1961. Cybernetics, or Control and Communication in the Animal and the Machine, Znd ed.
Cambridge, Mass.: MIT Press. Original work published in 1948.
- - . 1964. Ex-prodigy. Cambridge, Mass.: MIT Press.
Wilensky, R. 1983. Planning and Understanding: A Computational Approach to Human Rta5oning.
Reading, Mass.: Addison-Wesley.
Wilson, M. D., ed. 1969. The Essential Descartes. New York: Mentor.
Winograd, T. 1972.. Understanding Natural Language. New York: Academic Press.
- - - . 1975. "Frame Representation and the Declarative-Procedural Controversy." In D. G.
Bobrow and A. Collins, eds., Representation and Undnstanding: Studies in Cognitive Science. New
York: Academic Press.
Winston, P. H. 1977. Artifidallntelligenct. Reading, Mass.: Addison-Wesley.
---;and Brown, R. H., eds. 1979. Artificial Intelligence: An MIT Perspective, vol. 2.: Underslanding
Vision, Manipulation, Computer Design, Symbol Manipulation. Cambridge, Mass.: MIT Press.
Witherspoon, G. J. 1971. "Navajo Categories of Objects at Rest," American Anlhropologist 73:
110-17.
Wittgenstein, L. 1961. Tradalus logico-philosophicus, trans. by D. F. Pears and B. F. McGuinness.
London: Routledge & Kegan Paul. Original work published in 192.1.
- - - . 1968. Philosophical Investigations, trans. by G. E. M. Anscombe. New York: Macmillan.
Wolff, R. P., ed. 1967. Kant. New York: Doubleday.
Woods, B. T. 1980. "Observations on the Neurological Basis for Initial Language." In D.
Caplan, ed., Biological Studies of Mental Processes. Cambridge, Mass.: MIT Press.
Woods, W. A. 1975. "What's in a Link: Foundations for Semantic Networks." In D. G.
Bobrow and A. Collins, eds., Representation and Underslanding: Studies in Cognitive Science. New
York: Academic Press.
Young, R. M. 1974. "Production Systems as Models of Cognitive Development." In Proceedings
of AISB Summer Conference. University of Sussex.
Zaidel, E. 1977. "Lexical Organization in the Right Hemisphere." In P. Busser and A. Rougel-
Buser, eds., Cerebral Correlales of Conscious Experience. Amsterdam: Elsevier.
---.1978. "Auditory Language Comprehension in the Right Hemisphere Following Cere-
bral Commissurotomy and Hemispherectomy: A Comparison with Child Language and
Aphasia." In A. Caramazza and E. B. Zurif, eds., Language Acquisition and Liznguagt Breakdown:
Parallels and Dim-gencits. Baltimore: Johns Hopkins University Press.
Zangwill, 0. L. 1980. "Kenneth Craik: The Man and His Work" (The Kenneth Craik Lecture,
10 March 1978), British journal of Psychology 71: 1-16.
408
NAME INDEX
Ach, Narziss, 106 Bloomfield, Leonard, 196, 202-5, 208, 232,

Adler, Jonathan, xiv, 377 250, 292
Allport, Allan, .riu, 123 Boas, Franz, 204, 231-34, 235, 241, 254, 292
Anderson, James, 167 Bobrow, Daniel, xiv, 152, 162, 178
Anderson, John, xiu, 131-33, 136, 179, 331- Boden, Margaret, .riu, 164-65
32,368 Bohr, Niels, 60
Angell, James, 109 Boole, George, 60, 143-44, 150
Appel, Dolly, xu Boring, E. G., 291
Aristotle, 4, 16, 60, 62, 291, 323, 340, 342, Bousfield, John, 349
353, 360, 363, 366 Brandes, Stanley, .riu
Armstrong, Sharon, 347 Bransford, John, 124-25
Ashby, W. Ross, 25-26 Brentano, Franz, 101, 102, 105
Atkins, John, 246-49, 250 Bresnan, Joan, xiu, 217-18, 220
Atkinson, Richard, 122 Brewster, David, 49
Austin, George, 93 Broadbent, Donald, 91-93, 95, 119-20, 122-
Austin, J. L., 69-71, 77 23, 132
Ayer, Alfred J., 63, 70 Broca, Paul, 267, 268, 284
Brown, John Seely, xiu, 178
Brown, Roger, xiu, 343, 344
Brownell, Hiram, xiv
Brugmann, K., 197
Bruner, Jerome, .riu, 23, 24, 25, 29, 32, 64,
Babbage, Charles, 142-43 93-95, 110, 119, 126, 132, 140, 195, 293,
Ballard, Dana, .riu 341, 354, 361, 382
Balzell, C. E., 190 Bryant, Peter, xiu
Bartlett, Frederic, 26, 114-16, 119, 125-26 Buhler, Karl, 106
Barwise, Jon, xiu Bundy, McGeorge, 32
Bates, Elizabeth, xiu Bush, Vannevar, 20, 144
Bateson, Gregory, 24, 27
Bayes, Thomas, 374
Bell, Charles, 266
Benedict, Ruth, 234
Beneke, Friedrich Eduard, 99 Campbell, Donald, 254
Berkeley, George, 53, 54, 55-56, 57, 61, 87, Cantor, George, 60
105 Caramazza, Alfonso, .riu
Berlin, Brent, .riu, 252-53, 255, 294, 348-50, Carbone, Linda, xu
358 Carnap, Rudolf, 23, 63-65, 70, 84
Berlin, Isaiah, 70 Carroll, Lewis, 183
Bernstein, Alex, 139 Cazeneuve, Jean, 225
Bigelow, Julian, 15 Chapin, Damaris, xu
Binet, Alfred, 116 Cherniak, Christopher, 378-79
Bloch, Bernard, 190 Cherry, Colin, 91-93, 95
Block, Ned, xiv, 336 Chomsky, Noam, .riv, 23, 25, 28, 29, 34, 64,
409
Name Index
Chomsky (confinutd} Fodor, Jerry, riv, 34, 65, 81-86, 132, 215, 242,
75, 81, 84, 86, 125, 140, 168, 170, 182-96, 312-13, 314, 315, 332, 387, 391
199-200, 202, 205, 207-28, 239, 241, 242, Foucault, Michel, 291
253,264,293,306,312,314,332,382,386, Franz, Shepherd Ivory, 260--61
391 Frazer, Sir James, 229-30
Cohen, Gillian, xiu Frege, Gottlob, 16, 23, 60, 61, 227, 292, 360
Cohen, L. Jonathan, 373--78 Freud, Sigmund, 15, 75, 166, 196, 380, 389
Colby, Kenneth, 156-57, 158, 159, 161 Fries, Charles, 204
Cole, Michael, riv, 254, 256 Fries, Jakob Friedrich, 99
Conklin, Harold, 30 Fritsch, Gustav Theodor, 267, 284
Copernicus, Nicholas, 59
Craik, Kenneth, 16, 293
Galanter, Eugene, 32
Galileo, 75
D'Andrade, Roy,riu, 34,248, 25~57, 369-70 Gall, Francis Joseph, 266, 267, 278, 284
Darwin, Charles, 8, 14, 38, 227, 228, 291, 392 Galvani, Luigi, 266
Dejerine, Jules, 267--68 Gardner, Hilde Weilheimer, u
Democritus, 296 Gardner, John, 32
Dennett, Daniel, riu, 79-81, 82, 160, 178 Gardner, Ralph, v
Descartes, Rene, 4, 43, 49-57, 66--67, 72, 73, Gazdar, Gerald, 217-18, 220
74,75,81,82,85,86,87,142,192,196,216, Gazzaniga, Michael, riu
224, 238, 265, 278, 284, 291, 323, 392 Geertz, Clifford, riv, 243--44, 250, 253, 258,
Dewey, John, 73, 77, 109 355-56, 358, 359, 380, 388
Dodwell, Peter, 74, 85, 86 Gerard, Ralph, 23
Donders, F. C., 101, 102, 121, 132 Geschwind, Norman, riu, 275
Dresher, Elan, 177 Gibson, James J., 118, 294, 296, 307, 308-16,
Dreyfus, Hubert, riv, 81, 163, 164, 387 319, 320, 321, 322, 332, 346, 358
Duncker, Karl, 113 Gilson, Etienne, 49
Durant, Will, 56 Glass, Arnold, 134
Gleitman, Henry, 347
Gleitman, Lila, 347
Glucksberg, Samuel, riv
Goldstein, Kurt, 268
Goodenough, Ward, 30, 245, 248
Ebbinghaus, Hermann, 104-5, 115, 116, 126- Goodman, Nelson, riu, 64, 65, 351
27 Goodnow, Jacqueline, 29, 93
Eccles, Isabel, xu Gregory, Richard L., 269
Edgerton, Robert B., 246 Greissman, Judith, xu
Ehrenfels, Christoph von, 111 Grimm, Jakob, 197, 198
Einstein, Albert, 60, 113 Grodzinsky, Josef, riu
Euclid, 65, 139 Gross, Charles, riu
Evans, T. G., 151-52
Haddon, A. C., 230

Farah, Martha, riu, 331 Halle, Morris, 201
Fechner, Gustav, 101, 102, 106 Halstead, Ward, 272
Feigenbaum, Edward, 34, 155-56, 160--61 Harris, Marvin, 238
Feigl, Herbert, 63 Harris, Roy, 221, 388
Feldman, Jerome, riu Harris, Zellig S., 186, 188, 206-7
Feldman, Julian, 34 Harth, Eric, 283
Ferrier, David, 267 Hayes, Patrick, riu, 155
Fillmore, Charles, riv Head, Henry, 116, 268
Finke, Ronald, 331 Hebb, Donald, 26, 271-72, 320
Flourens, Pierre Jean Marie, 266--67, 268 Heidegger, Martin, 73
410
Name Index
Heider, Eleanor, see Rosch, Eleanor Keesing, Roger, 252
Helmholtz, Hermann von, 99-102, 117, 292, Kessen, William, 130
296, 305, 307, 308 Kessler, Martin, ru
Herbart, Johann Friedrich, 99 Keyser, Samuel Jay, riu
Herodotus, 226 Klivington, Kenneth, riii
Herskovits, Melville, 254 Kliiver, Heinrich, 23, 272
Hinton, Geoffrey, riu, 167, 387 Koffka, Kurt, 111
Hitzig, Eduard, 267, 284 Kohler, Wolfgang, 111, 112-14, 271, 272-73
Hjelmslev, Louis, 196 Kortz, Nan, rv
Hockett, Charles F., 204, 206, 208 Kosslyn, Stephen, riv, 129, 294, 325, 326-36,
Hofstadter, Douglas, 176 330-32, 366, 382
Holyoak, Keith, 131, 134 Krechevsky, Mara, rv
Hornstein, Norbert, 177 Kripke, Saul, 351-52, 353, 360
Hoss, Phoebe, ru Kroeber, Alfred, 234
Huarte, Juan, 265 Kuhn, Thomas, 209
Hubel, David, 30, 273-74, 282, 283, 284-85, Kiilpe, Oswald, 105, 106
302, 304, 307 Kyburg, Henry, 378
Hull, Clark, 110
Hume, David, 53, 54, 55-56, 57, 61, 81, 87,
291
Hunt, Morton, 36
Hutchins, Edwin, 256
Lakoff, George, riv, 209, 212
Langness, L. L., 246
Lashley, Karl, 11, 12-14, 15, 20, 110, 118,
186, 191, 192-93, 260-64, 268, 270, 272,
273, 276, 278, 283, 382
lnhelder, Barbel, 377 Lave, Jean, riu, 256
Isard, Stephen, riu Leach, Edmund, 241
lsay, Jane, ru Leary, David, 99
Lederberg, Joshua, 155
Lees, Robert, 186, 189-90
Leibnitz, Gottfried, 57
Lenat, Douglas, 171
Lenneberg, Eric, 343, 344
Jackson, Hughlings, 269 Lettvin, Jerome, riu, 30, 274, 307
Jakobson, Roman, 26, 189, 196, 200--2, 205, Levine, Linda, riv
219, 221, 235-36, 245 LeVine, Robert, riv
James, William, 107-9, 292 Levi-Strauss, Claude, riu, 26, 75, 140, 195,
Johnson, Samuel, ru 202, 235, 236-44, 257-59, 293, 350
Johnson-Laird, Philip, riu, 294, 331-32, 361- Levy-Bruhl, Lucien, 223-25, 257-59
70, 373, 379, 380, 382 Lighthill, Sir James, 164
Jones, Sir William, 196-97 Locke, John, 4, 53, 54-55, 57, 61, 71, 72, 73,
Jung, Carl, 380 77, 83, 87, 105, 291
Loffredo, Carmella, ru
Longuet-Higgins, Christopher, riu, 171, 179-
80, 306
Lorenz, Konrad, 31
Lounsbury, Floyd, 30, 245, 248
Kagan, Jerome, riv Lowie, Robert, 228
Kahneman, Daniel, 361, 370-79, 400 Luchins, Abraham, 113
Kandel, Eric, 279-80, 281-82 Luria, Alexander, 16, 26, 276
Kant, Immanuel, 4, 53, 56-60, 61, 65, 68, 71,
72,73,75,76,86,87,98-99, 100,114,116,
117, 118, 137, 192, 291, 392
Kaplan, Ronald, riv
Kardiner, Abram, 229
Katz, Jerrold, 34, 215 McCarthy, John, riu, 23, 30, 138-39, 145,
Kay, Paul, riu, 252-53, 255, 294, 348-49, 154-55, 178
350, 358 McClelland, Jay, riv
411
Name Index
McConnell, Susan, .riv Norman, Donald, .riu, 36, 387
McCorduck, Pamela, 138 Nottebohm, Fernando, 2.80-82.
McCulloch, Warren, 10,18-19, 2.0, 2.1, 2.2., 2.4,
2.6, 30, 139, 144, 148, 2.72.-73, 307, 32.0
McDougall, William, 118, 2.30
Magendie, Fran.;ois, 2.66
Malcolm, Norman, 74 Olivier, Donald, 344
Malinowski, Bronislaw, 2.33-34, 2.92. Oppenheimer, }. Robert, 2.4, 2.5, 95
Mandler, Jean, riv, 2.9 Osherson, Daniel, riu, 347-48, 350
Marbe, Karl, 106 Osthoff, H., 197
Marcus, Mitchell, 169
Marie, Pierre, 2.68
Marler, Peter, 2.81
Marr, David, 170-71, 2.16, 2.94, 2.98-308, 310,
311, 319, 32.0, 32.1-2.2., 382., 384 Paivio, Allan, 330
Marshack, Alexander, riu Panini, 196
Marshall, John, xiv, 142., 179, 180, 2.82. Papert, Seymour, 169
Marx, Karl, 75 Parisi, Domenico, .riu
Maturana, Humberto, 30 Paul, Hermann, 197
Mehler, Jacques, xiu Pavlov, Ivan, 12.
Messer, August, 106 Peano, Cduseppe, 60
Metrodorus, 2.96 Peirce, Charles Sanders, 60
Mettrie, Julian Offray de Ia, 142. Peters, Stanley, riu
Metzler, Jacqueline, 12.9, 32.5 Piaget, Jean, 2.6, 75, 83,116-18,119,132.,136,
Meyer, Christine, .riv 140, 2.14, 2.19, 2.42., 2.54, 2.64, 2.92., 377,
Meyers, C. S., 2.30 389
Millar, Susanna, .riu Pinker, Steven, 2.15, 330
Miller, George, .riv, 2.3, 2.4, 2.6, 2.8-2.9, 32., 64, Pitts, Walter, 18-19, 2.0, 2.1, 2.6, 144, 32.0
89-91, 92., 95, 97, 110, 116, 119, 132., 140, Planck, Max, 60
195, 2.14, 2.39, 2.64, 2.93, 382. Plato, 3, 4, 52., 76, 77, 86, 191, 192., 2.12., 2.91,
Minsky, Marvin, .riv, 2.5, 2.6, 30, 34, 139, 140, 2.96, 358, 392.
145, 151-54, 161, 165-66, 167, 169 Popper, Karl, 76-77
Mishkin, Mortimer, 2.74 Posner, Michael, .riv, 2.9, 32.
Montague, Richard, 64, 2.16 Preble, Edward, 2.2.9
Montesquieu, Charles Louis, 2.2.6-2.7, 2.2.8, Pribram, Karl, .riv, 32., 2.64, 2.70, 2.83
2.41 Putnam, Hilary, .riv, 31, 66, 77-78, 81, 82., 88,
Moravcsik, Julius, riv 195, 351-52., 353, 358, 388
Moray, Neville, 12.0 Pylyshyn, Zenon, 175, 2.94, 312.-13, 314, 315,
Morgan, Lewis Henry, 2.45 332.-36, 368, 387, 391
Morris, Charles, 2.45
Morton, John, riu, 134, 2.82.
Moses, Joel, 161
Mountcastle, Vernon, 2.74
Muller, Johannes, 2.66 Quillian, Ross, 169
Murray, Stephen, 2.52.-53 Quine, Willard V. 0., 65, 66, 70-71, 74, 195,
351
Neisser, Ulric, .riu, 33, 114, 12.0, 133-34, 318,

331
Neurath, Otto, 63 Radcliffe-Brown, A. R., 2.33, 2.37, 2.92.
Newcombe, Freda, .riv Reddy, Raj, xiu, 169
Newell, Allen, .riu, 2.3, 2.5, 2.6, 2.8, 2.9, 30, 34, Rey, Georges, 336
64, 139, 140, 141, 145-51, 154, 155, 160, Rivers, W. H., 2.30
161, 179,2.19,2.93,319,32.0,356,361,382., Rochester, Nathan, 139
385, 386 Romney, Kimball, 34, 2.48
Newmeyer, Frederic, 190, 195, 2.06, 2.12. Rorty, Richard, .riu, 71-76, 77, 82., 85, 86, 87,
Newton, Isaac, 38 2.82.
Nielsen, J. M., 2.72. Rosch, Eleanor, .riv, 2.55, 2.94, 342., 344-48,
Nishahara, Keith, 304 349, 350, 358, 382.
412
Name Index
Rosenblatt, Frank, 307 Talmy, Leonard, .riu
Ross, J. R., 209 Tarski, Alfred, 65
Rousseau, Jean-Jacques, 241 Thorndike, Edward L., 12, 110
Rumelhart, David, .riv, 125 Tinbergen, Niko, 31
Russell, Bertrand, 17, 23, 60, 61, 62, 64, 65, Titchener, Edward Bradford, 105, 107,
73, 143, 145-46, 291, 360 108
Ryle, Gilbert, 66--68, 69, 70, 71, 80, 81, Tolman, Edward C., 110, 118
82 Treisman, Anne, 120
Troubetskoy, Nikolay, 200
Turing, Alan, 16, 17-18, 25, 26, 139, 144, 156,
188, 293, 382, 385
Turvey, Michael, 314, 315, 316
Sahlins, Marshall, 349-50, 356 Tversky, Amos, 294, 360, 361, 370-79, 389
Samuel, Arthur, 139 Tyler, Stephen, 34, 245, 251
Santa, John, 134 Tylor, Edward, 227-29, 231, 256, 292
Sapir, Edward, 183, 195, 204, 205, 234-35
Saussure, Ferdinand de, 196, 198-200, 202,
219, 221, 232, 236, 292
Schank, Roger, riu, 23, 165, 167-68, 177-78,
216
Scheffler, Israel, riv Ullman, Shiman, 311-12, 314, 321
Schlick, Morris, 63
Schwartz, Stephen, 352
Scribner, Sylvia, 256
Searle, John, .riu, 81, 140, 168, 171, 172-76,
177, 199, 249
Segall, Marshall, 254 van Lehn, Kurt, 178
Shannon, Claude, 21, 22, 144, 188, 382 Vaucanson, Jacques de, 142
Shanon, Benny, xi~ 123,388 Vico, Giambattista, 4
Shaw, Cliff, 146, 147 Voegelin, C. F., 190
Shaw, George Bernard, 229 von Ehrenfels, ste Ehrenfels, Christoph von
Shaw, Robert, 314, 315, 316 von Helmholtz, Ste Helmholtz, Hermann von
Shebar, William, 336-38 von Humboldt, Wilhelm, 196
Shepard, Roger, 128--30, 132, 324-25, 330, von Neumann, John, 10, 17, 18, 19, 21, 22, 23,
338 24,26,29,139,145,180,188,307,319,320,
Shiffrin, Richard, 122-23 382, 385
Shulman, Gordon, 32 Vygotsky, Lev, 26
Siegler, Robert, riv
Silverstein, Michael, 250
Simon, Herbert, .riu, 22, 23, 25, 26, 28, 29, 30,
33, 34, 64, 139, 140, 145-51, 154, 155, 160,
161, 179, 219, 241, 264, 293, 319, 320, 356, Wagner, Sheldon, riu
361, 382, 385, 386 Wallace, Anthony, 246-49, 250
Simon, Theodore, 116 Waltz, David, 170, 297
Skinner, B. F., 12, 80, 110, 191-93, 194, 310, Wanner, Eric, riii
312, 314 Wason, Peter, 361-62, 369, 373, 379
Sloman, Aaron, .riu, 138 Watson, John B., 12, 109-11, 260, 264, 292
Slovic, P., 373 Watt, Henry, 106
Smith, Brian Cantwell, .riu Weaver, Warren, 21
Smith, Edward, 347-48, 350 Weiss, A. P., 203
Socrates, 3, 4 Weiss, Paul, 23, 270
Spence, Kenneth, 110 Weizenbaum, Joseph, 156-57, 158, 162, 163,
Sperber, Daniel, 242-43, 380 177
Sperling, George, 120 Wernicke, Carl, 267, 284
Sperry, Roger, 270, 275-76 Wertheimer, Max, 111-12, 113, 114
Stephens-Swannie, Laura, xu Wexler, Kenneth, 177
Sternberg, Saul, 121-23 Whately, Bishop, 227
Strauss, Claudia, riu White, Leslie, 233
Sussman, Gerald, 160 Whitehead, Alfred North, 3, 17, 23, 52-53,
Sutherland, Stuart, riv, 307 60, 61, 62, 64, 143, 145-46, 291, 360
413
Name Index
Whorf, Benjamin, 205, 2.32., 2.35, 2.55, 344, Wolf, Connie, ru
357,358 Woods, William, 169
Wiener, Norbert, 15, 2.0, 2.1, 2.2., 2.3, 2.4, 2.6, Wundt, Wilhelm, 102.--4, 105-7, 108, 117,
115, 139, 144, 188, 2.93, 2.95, 307 136, 198, 203, 2.2.7, 2.92., 32.4
Wiesel, Torsten, 30, 2.73-74, 2.82., 2.83, 2.84-
85, 302., 304, 307
Wilensky, Robert, 141
Winner, Ellen, .riu
Winograd, Terry, Iiu, 158-60, 161, 162., 164,
165, 177, 178, 2.16, 2.97 Young, Richard, 171
Winston, Patrick, 170
Witherspoon, Gary, 2.50
Wittgenstein, Ludwig, 62., 63, 68-70, 71, 73,
74, 76, 77, 2.86-87, 336-39, 342., 346, 351,
353, 357, 358 Zurif, Edgar, .riu
414
SUBJECT INDEX
ablation: Karl Lashley's use of, 261; tech- 78; Dartmouth College meeting, 30, 138-
nique, 261, 273; of the visual cortex, 261 40; definition of, 140; development of,
Adaptive Control of Thought (ACf), 131- 138-39; and intentional systems, 80; and
32; process of, 131 knowledge, 161; and language, 167-69,
adequacy, Noam Chomsky's standards of, 216-17; and MPPS, 319; and perception,
189 169-71, 304, 319; and problem solving,
adhesion, 228 160-61; and psychology, 135,137, 180-81;
affordances, 310, 312, 315, 317 and reasoning, 379; SHRDLU, 158-60;
agnosia, 22 tensions in, 141; top-down approach, 165-
Alfred P. Sloan Foundation, riii, 35-38, 220; 66; see also computers, information process-
SOAP, 36-37 ing
algorithmic level (David Marr), 299-300 Asptcls of tht Thtory of Synftlx (Chomsky),
American Psychological Association, 97 207-8; responses to, 209
anthropology,26,223-59,341,350,354,356, assemblers, 145
379; in America, 213; case studies, 225; and avunculate, 236-37, 240
classification, 341, 355-56, 358; cognitive,
see cognitive anthropology; and cognitive
science, 41, 44; and color terms, 255; and
computers, 243-44; definition of culture,
228; ethnoscience, 225; functionalism, 227, baseline (or base) rates, 374
233-34; and illusion studies, 254; and lan- basic level, 346
guage,234-36,255;andlearning,228;and Bayes theorem, and the law of statistics,
linguistics, 202, 222, 232, 234-36; and 374-76
mathematical skills, 256; methodology in, behaviorism, 10--16; 23, 26, 65, 86, 94, 109-
228; myths, 239-41; and primitive cul- 11, 119, 191-92, 193, 214, 281, 282, 292-
tures, 223-25, 230, 238-39; and primitive 93, 323,324, 381-82; in America, 113, 292-
mentality, 223, 230, 232; and psychology, 93; and classification, 94; and cognitive
230, 254; and religion, 229; scientific, 231; science, 381; and computers, 10--11; and
symbolic, 355-56; and symbolic processes, the cybernetic approach, 32; differences in
243-44; and transformationalism, 242; and societies, 226; and ethology, 31; and Ge-
visual perception, 254; see also componen- stalt psychology, 113; and introspection,
tial analysis, ethnoscience, ethnosemantic 11-12, 109; in linguistics, 191-92, 203-4,
approach, kinship, structural anthropol- 214; and localization, 270; and logical em-
ogy piricism, 67; in mental imagery, 323; and
Antipodeans, 74, 75, 78 neuroscience, 261, 262, 265; in psychol-
aphasia, 22; and localization, 267 ogy, 109-11, 130, 135, 260; and represen-
Aplysia mli/ornica, 279-80, 282, 287 tational level, 39; revolution, 109; of a sys-
apperception, 103 tem, 332; see also reductionism
artificial intelligence, 5, 40-41, 42, 44, 54, 80, Bible, 226
87, 130, 131, 138-81, 194, 216, 217, 298, bifurcation, 248
304-5,306,379, 388; beginnings of, 142; in bird songs, 280--82
Britain, 164; and cognitive science, 40; and bit (binary digit), 21
computers, 41; critics of, 81, 162-65, 177- bottom-up approach: in Gestalt psychology,
415
Subject Index
bottom-up approach (continued} cognitive anthropology, 30, 244, 245, 250,
112; and David Marr, 305; in psychology, 252; and language, 250-51
97 cognitively penetrable, 333
brain, 31, 264, 275, 332, 384; and forms of Cognifivt Psychology (Neisser), 33, 133
behavior, 284; as a holographic process, cognitive psychology, 42, 44, 95-98, 119-20,
283-84; and perception, 112,300,306, 307; 130; and Adaptive Control of Thought
see also hemispheres, mind (ACT), 131; and artificial intelligence, 180-
brain injury, 269, 277-78, 331; and the im- 81; and bottom-up approach, 97, 126; fu-
age-generation component, 331; and local- ture of, 134; and Gestalt psychology, 114-
ization, 267; and the nervous system, 284 15; history of, 119; and information
Brain Mechanisms and lnfelligmce (Lashley), 261 processing,95-96,119-21,130-34;andin-
formation theory, 96; and memory, 122,
131 (set also memory); and mental repre-
sentations, 128; methodology of, 97-98;
and modular analysis, 132-33; molecular
calculator, 142 vs. molar, 97; problems in, 96; top-down
California Institute of Technology, 10 approach, 97, 124-28; set also psychology
Cambridge Anthropological Expedition, 230, cognitive science: critics of, 42; definition of,
253 6; representations, 38; symptoms, 38
Carnegie Corporation, 32 color: componential analysis, 251; lexicons,
case studies, 225 349
categories of thought, 58 color naming, 324, 343, 344-50; and classical
categorization, 340-59; and anthropology, view of categories, 342; in the Dani, 344;
341, 355-56, 358; boundaries of, 346; and and nervous system, 348
the classical view, 341-42, 345, 347-48; color terms, 255, 258; basic, 350
and cognitive science, 358-59; and color compilers, 145
lexicons, 349; and color naming, 342-50; componential analysis, 244, 246-51, 253,
critics of classical view, 345, 347-48; and 257-58; and kinship, 246-49
empiricism, 342; and ethnoscience, 358; computation, 16-18
extensions of, 342; intensions of, 342; and computational paradox, 9, 44, 133, 180,
language, 351, 354, 358; and logic, 351; 384-88
natural view of, 341, 347, 348; and the ner- computatiopal theory, 299
vous system, 341, 348; and perception, Computer anatht Brain, Tht (von Neumann), 29
349-50, 354, 358, 359; and philosophy, computer languages, 369; LISP, 154-55, 161;
340, 341-42, 350-55; prototypes, 346; and machine, 146
psychology, 344-48; and scientists, 357; computer programs, 138, 166, 256, 329, 334,
structure of, 346-47; and symbolic anthro- 366-67; DENDRAL, 155-56, 161; ELIZA,
pology, 355-56; and Whorf-Sapir hypoth- 157; T. G. Evans, 151-52; GPS, 34, 148-51,
esis, 343 154; HACKER, 160; Logic Theorist (LT),
cells: complex, 273; hypercomplex, 273; sim- 145-48; STUDENT, 152-53, 157; Turing
ple, 273 machine, 18; on visual analysis, 152, 299
Center for Advanced Study in the Behavioral computers, 5, 12, 15, 29, 40-41, 87, 95, 118-
Sciences (Stanford University), 25 24, 134, 145-51, 171, 172, 300, 328-29,
Center for Cognitive Studies (Harvard Uni- 333, 354, 361, 366-67, 379-80, 381, 384-
versity), 32 88; analog, 20; assemblers, 188; and
Center for the Study of Language and Infor- behaviorism, 10-11; in Britain, 16; and
mation (Stanford University), 217, 218 cognitive science, 40-41, 381, 382, 384-88;
centralists, 133 and data structures, 328-29; difference
Cerebral Mechanisms in Behavior Confer- machines, 142; digital, 44, 130, 167, 180,
ence, set Hixon Symposium 264, 320, 325, 334, 385; electronic, 6, 10,
children: and learnability theory, 215; Jean 21, 321; and functionalism, 78; and inten-
Piaget's study of, 116-17, 377; and syllo- tional systems, 79-81; Johnniac, 146; list
gisms, 366 processing, 146; and logic, 361, 366-67;
Chinese Room problem (John Searle), 172- and memory, 30, 34,119, 121-22, 146; and
74, 249; critiques of, 173-74 mental imagery, 328-29, 332, 334; and
classification, 238-39, 292; and behaviorism, mental representation, 328-29; and the
94; classical features of, 341-42; and em- model of the mind, 6, 10, 40, 41, 44, 82,
piricism, 56; and psychology, 93-95 148, 166, 264, 320, 331, 332, 335, 354, 380,
Cognition and Reality (Neisser), 134 388; and perception, 297-98, 300, 307; and
Cognition Project (Harvard University), 93 psychology, 118-24; and representations,
416
Subject Index
3~7; simulation of, 2.45, 32.8-30; space 61--62.; and mental representations, 82.-83;
allocation for, 146; symbolic form of. 34, and neurophysiology, 75-76; in philoso-
12.9, 149; and John von Neumann, 2.9, 145, phy, 49, 53, 60, 71-81; and psychology,
180, 319, 32.0, 385; set also artificial intelli- 74-75; and rationalism, 81; see also knowl-
gence, information processing, informa- edge, philosophy
tion theory equipotentiality, 2.62., 2.70
computer science, 91, 134, 2.72. Eskimos, Franz Boas's fieldwork, 2.31
concept formation, 12.6 Establishment (in perception), 313-18; and
conceptual systems, 2.55 Jerry Fodor, 313-15; Gibsonians' attack of,
connotations, 2.50 316-18; and Zenon Pylyshyn, 313-15; rep-
Critique of Pure Reason (Kant), 43, 57, 59, 392. resentations, 313; and Robert Shaw, 314-
culture, 2.56-57, 343; and color naming, 343, 15, 316; terminology, 315, 317; and Mi-
349; and language, 2.35-36; linear evolu- chael Turvey, 314-15, 316
tion of, 2.31; Edward Tylor's definition of, ethnography, 2.52.
2.2.8 ethnoscience, 2.2.5-2.6, 2.44-53, 355; and cate-
Cybernetics (Wiener), 2.0 gories, 358; and cognitive anthropology,
cybernetics, 19-2.1, 2.4, 32., 144-45, 2.43, 2.45; 2.44, 2.45; and componential analysis, 2.44,
conference on, 2.4; and Norbert Wiener, 2.46-49, 2.51; decline in, 2.52.-53; and eth-
2.0, 144-45 nosemantics, 2.44; goals of, 2.46; and kin-
ship, 2.46-49; and linguistics, 2.46; research
in, 2.52.
ethnosemantic approach, 30, 2.44, 2.46, 2.52.-
53, 2.58
Dani, color studies, 344 ethology, 31
Dartmouth College Meeting, 30, 138-40, Euclid's principles, 65, 139
145-58 evolution, 2.31, 2.32., 2.33; linear, 2.31
declarative representations, 161--62. external properties, 332.
deep structure, 2.07-8, 2.09, 2.12.
DENDRAL, 155-56, 161
determining tendency, 106
difference machine, 142.
direct perception, 311, 312. field theory, 2.71
Discourst on Method, The (Descartes), 392. finite-state grammar, 185, 186, 193
0-structure, 2.18 Ford Foundation, 2.5
foundational philosophy, 56-59; and "cate-
gories of thought," 58; and Immanuel
Kant, 56-59; and knowledge, 57-59; and
the mind, 57; and psychology, 59; and rep-
ecological approach, 315, 316, 317-18; and resentation, 59; and schemas, 58-59; and
the Establishment, 314-18; of James }. the sensory world, 58; and synthetic a priori,
Gibson, 314-15; of Ulric Neisser, 318 57
effective procedure, 367 functional architecture, 333
effectivities, 315, 317 functionalism, 107-9, 162.
ELIZA, 157, 162. functionalism (philosophical), 53, 78-79,
empiricism, 8, 53-.56, 57, 62--63, 81-82., 86, 251, 2.92.; in anthropology, 2.2.7, 2.33-34;
88, 192., 195, 2.14-16, 2.91, 317, 361; agen- and computers, 78; and epistemology, 82.;
da, 56; a priori, 57; British, 342.; and classifi- and the mind-body problem, 78-79
cation, 56, 342.; and introspection, 54, 58, fuzzy set logic, 347
60; and language, 56; and molar analysis,
97; and perception, 55, 309; and philoso-
phy, 52., 54, 353; and psychology, 71, 101;
and Vienna Circle, 65; see also logical em-
piricism General Problem Solver (GPS), 34, 148-51,
engram, 2.62--63 154, 179, 32.0
epistemology, 19, 60, 61--62., 70, 71-81, 81- generative grammar, 187, 194, 2.10, 2.16
86, 87, 116-18, 382.; agenda, 62.; and artifi- Gestalt psychology, 111-18, 119, 12.6, 12.8,
cial intelligence, 31; critics of, 86-87; and 12.9,2.62.,2.65,2.70,271,2.76,2.92.,32.1;anal-
cultural relativism, 77; and empiricism, 81; ysis of, 112.; and artificial intelligence, 113;
and functionalism, 82.; genetic, 117-18; is- and behaviorism, 113; and holism, 112.,
sues in, 14; and language, 82.-84; and logic, 114, 2.69, 2.70-71; and learning, 113; and
417
Subject Index
Gestalt psychology (confinutd) inference, 100; and philosophy, 62
localization, 271; and logical formalisms, information processing: 51, 90-91, 119-20,
117-18; and memory, 115-16; and move- 121,123,125,129,132,134,148,193,239,
ment, 111-12; and perception, 111-12, 305, 307, 311, 318; and architecture, 167;
114,199, 296; and schemas, 116; and struc- Atkinson-Shiffrin model, 122-23, 124-25;
turalism, 113 in Britain, 91-93; and the classical model,
Golden Bough, The (Frazer), 229 128; and cognitive psychology, 95-96, 134;
grammar, 206, 221; finite-state, 185, 186, and computers, 325; criticism on approach,
193; generative, 187, 194, 210, 216; lexi- 133; devices, 132; early systems, 119-21;
cal-functional, 217; phrase-structure, 186; functional architecture, 333; and language,
story, 125; transformational, 186-88, 215, 124-25; limitations, 90; and memory, 121-
216, 218; universal (U.G.), 211-12 24; and mental representations, 128-30;
Greeks, 3-5, 7, 14, 50, 53, 75, 87, 324, 340, and modular analysis, 132-33; and psy-
356, 363, 381, 385; on behavior, 265; and chology, 90-91, 119-21, 122-24, 130-32,
cognitive science, 42.-43; on definition of 311; systems, 40; top-down perspective,
man, 360; epistemology, 14; on grammar, 124-30; see also artificial intelligence, com-
196; on knowledge, 4-5, 291, 296; on per- puters
ception, 296; philosophers, 294; on ratio- information theory, 21-22, 119, 128, 243,
nality, 294 245; and Colin Cherry, 91; and Oaude
Shannon, 21; and Norbert Wiener, 21
Institute for Advanced Study (Princeton
University), 24
intelligence: in apes, 113; in computers, 154;
and learning, 113
HACKER, 160 intentionalism, and philosophy, 53
Harvard University: Center for Cognitive intentional systems, 79-81; and artificial in-
Studies, 32; Cognition Project, 93; philoso- telligence, 80; and computers, 79; Daniel
phers at, 65; Society of Fellows, 25-26 Dennett on, 79-80; Zenon Pylyshyn on,
HEARSAY, 169 175-76; and John Searle, 175-76; Turing
hemispheres, 267-68, 275, 276-78, 284; in machine, 79
birds, 281; effects of injuries on, 277; func- internal representations, 337-38
tions of, 275; and holism, 275; and the im- International Business Machines (IBM), 139
age-generation component, 331; and lan- intrinsic properties, 332
guage, 276-78; left, 275, 276-78, 281, 331; introspection: and behaviorism, 11-12; and
right, 275, 276-78, 281; see also neuro- cognitive psychology, 101, 119; and em-
science piricism, 54; and Hixon Symposium, 11-
Hixon Symposium, 10-16, 18, 23, 29, 110, 12; in philosophy, 50; and Edward Brad-
119, 264, 278, 293, 361, 382; and behavio- ford Titchener, 107; and Wilhelm Wundt,
rism, 11-12; importance of, 25; interdisci- 102-3, 107
plinary studies of, 42; and introspection, IQ test, 116; and Jean Piaget, 116
11-12; and neuroscience, 272-73
holism, 268-70, 271, 275-76, 278, 279, 284,
292; and Gestalt psychology, 268-69; and
Richard Gregory, 269; and hemispheres,
275; and Hughlings Jackson, 269; vs. local-
ization, 270; and nervous system, 268-70; Josiah P. Macy Foundation, 23-24, 36; Sym-
and psychology, 268-69 posium, 42, 293
holography, 283-84 junggrammatilcer, see neo-grammarians
human behavior, 171
human body (mechanical), 51
kernel sentences, 187, 207

kinship, 236-37, 246-49, 251; bifurcation,
illusions: Miiller-Lyer test, 254; studies in 248; and componential analysis, 246-49,
anthropology, 254 251; relations, 236; structures, 236-37; sys-
imagery, set mental imagery tems, 236
implementation (David Marr), 299 knowledge, 57-59, 60, 72-73, 79, 161; in ar-
individual ego, 57 tificial intelligence, 161, 178; and cognitive
418
Subject Index
science, 6; and empmc1sm, 54; and the America, 202; and anthropology, 202,232,
Greeks, 4-5, 291; origins of, 54; and phi- 234-36; and artificial intelligence, 216; and
losophy, 43, 52; representational systems, autonomy, 208; and behaviorism, 191-92,
162; study of, 298; see also epistemology 203-4; and cognitive science, 220; and col-
or naming, 344; diachronic, 198; history of,
197; and information processing, 193; and
leamability, 215, 217; and neo-grammari-
ans, 197-98; and philology, 198; and phi-
Language (Bloomfield), 204-5 losophy, 215-16; and phonemes, 26,
language, 14, 27, 62, 68, 69-70, 8.3--84, 104, 200-1, 236; and phonology, 200-2; and
168--69, 182-96, 209, 213, 217-18, 220, phylogeny, 197; and the Prague School,
221,222,235,251,287,292,306,314,336, 200-2; and procedures, 206; and psy-
339, 345, 354; and anthropology, 222, 232, cholinguistics, 214-15; and psychology,
234-36, 237-38, 250-51; and artificial in- 192, 193, 201, 214, 219-20; and rational-
telligence, 167-69, 177,216, 217; and brain ism, 379; and right hemisphere, 275; and
damage, 277-78; and categorization, 351, Standard Theory, 207; structuralism, 206,
354, 358; and cognitive anthropology, 208, 236; theory of, 194; and universal
250-51; comparative, 198; culture, 235-36; grammar, 210; see also Chomsky, language
debates in, 209; and deep structure, 207-8, LISP, 154-55, 161
209, 212; and empiricism, 55; families, 187; list processing, 146
and finite-state grammar, 185, 188, 193; as localization, 261, 262, 263, 265-72, 272-73,
a set of games, 68; and generative gram- 273-74, 275, 278, 284-85, 292; and behav-
mar, 187; and generative semanticists, 209; ior, 270; and the Hixon Symposium, 272;
in hemispheres, 267-68, 275-76, 277-78; vs. holism, 270, and language, 267-68; in
and interpretive semanticists, 209; and nervous system, 284-85; and plasticity,
kernel sentences, 187; learning of, 193, 284; and reductionism, 268; in visual sys-
278, 312; and localization, 267-68; and tem, 270
mental imagery, 336, 339; and neuro- logic, 16, 60, 61, 225, 331, 351, 360, 364-65,
science, 267-68, 276-78, 287; and nota- 367-70; capabilities of, 362; and computer
tions, 203; and philosophy, 69-70, 212; programs, 366-67; and computers, 361; de-
phonemes in, 27, 200-1, 236; and phrase- ductive processes of, 361; in history of
structure grammar, 186, 187, 188; and philosophy, 360; laws of formal, 365; ma-
primitive mentality, 232; and representa- chines, 363; and mental representations,
tional level, 39; and structuralists, 191, 368--69; and model of neuron, 361; and
195; and surface structures, 208, 209, 211; neural networks, 18; pre-logic, 223-24;
and transformational analysis, 186-87; and the primitive mind, 225; and psychol-
and transformational grammar, 187, 188; ogy, 361, 363-64, 370; and reasoning, 363,
and transformations, 187, 206-7; see also 369; in solving problems, 370-72; and syl-
linguistics, semantics, sentences, syntax logisms, 363-64, 365-66, 367-70; see also
Language of Thought, The (Fodor), 83 rationality, reasoning
langue, 196, 199 logical empiricism, 62-66, 67-71, 86-87, 292,
Latin, 196-97 351; criticism of, 66, 86-87; and language,
law of maximal contrast, 201 62-64, 65, 68, 69; and logical syntax, 64;
law of participation, 224 and the mind, 67; and physicalism, 63; and
law of probability, 372, 373, 374, 375 psychology, 68; verificationism, 63; and
learning, 132, 261-63, 271-72, 279-80, 287; the Vienna Circle, 63-65; see also empiri-
and anthropology, 228; in Aplysill californica, cism, philosophy
279-80, 282; and Gestalt psychology, 113; Logic Theorist (LT), 145-48
of language, 193; and language recovery, logic theory machine, 28
276; limitations, 90; and linguistic record-
ing, 90; and natural behavior, 282; after
surgical lesions, 261; theory, 215, 217, 282
Leipzig School, 107
lesion, 261, 268; technique of, 261
lexical-functional theory, 217 Macy Foundation, see Josiah P. Macy Foun-
lexical semantics, 220 dation
lexicons, 211, 212 "Magic Number Seven, Plus or Minus Two,
L 'Homme Machine (Mettrie), 142 The" (Miller), 29, 89
Lincoln Laboratories, 125 markedness, 202
linguistics, 44, 54, 90, 182-222, 232, 379; in Marxism, 233
419
Subject Index
massive parallel processing system (MPPS), "Minds, Brain, and Programs" (Searle),
319-21 171
mathematics, 17, 19, 50, 60-61, 300; and an- Mirror of Nature, set Philosophy and the Mirror
thropology, 256; and cognitive science, 16- of Nature
18; and logic, 61; Markov processes, 26 modal model, 122
melody, 111 modularists, 133
memory, 95, 115-16, 121-24, 125-28, 279; molar analysis, 100, 102, 116, 135; definition
and ACT, 131; Atkinson-Shiffrin model, of, 96-97; and empiricists, 97; and Gestalt
122; and brain damage, 262; and cognitive psychologists, 113
psychology, 122, 131; declarative, 131; and molecular analysis, 100, 101, 109, 135; defini-
functional architecture, 333; and Gestalt tion of, 96-97; and Gestalt psychologists,
psychology, 115-16; holographic theory 112; and rationalists, 97
of, 284; and information processing, 121- Moral Sciences Club, 76
24, 126; long-term, 122; and mental imag- movement: perception, 111-12; and Wert-
ery, 328--29; a model of, 121; and MPPS, heimer, 111-12
319; musical, 131; production, 131, 132; myths, 239-41; autochthony, 240; Oedipus,
and psychology, 121-24, 126-28, 131; 239, 240
schemas, 116; short-term, 28, 122-23;
stores, 122; working, 131; see also comput-
ers: memory
Mm11 (Plato), 3, 4, 43, 72, 83, 291, 392
mental imagery, 50, 323-39, 358; and be-
haviorism, 323; as cognitively penetrable, nativism, 194
333; and computers, 328--29, 332, 334; crit- natural-kind terms, 352-53
ics of Stephen Kosslyn's model of, 330-31; neo-grammarians, 197-98; and psychology,
and data structures, 328--29; flexibility, 198
326; and functional architecture, 333; and nerve-cell functioning, 279
the image generation component, 331; in- nervous system, 13, 19, 20, 260-63, 266, 267,
terest in, 324; and internal representations, 269,273-74,276,278-79,28Z-83;ablation
337-38; and Kosslyn's model and brain of, 261; and behavior, 10; and categories,
damage, 331; and language, 336, 339; and 341; and cells, 273; cognitive capacities of,
logic, 331-32; and map-location task, 326, 22, 41; and color naming, 348; and devel-
333; and memory, 328; and mental models, opment, 273-74; historical views of, 265-
332; and mental representations, 327, 330; 66; holism, 268--70, 271; and holography,
and nerve cells, 338; and neuroscience, 28Z-84; and lesion technique, 261; localiz-
330; and perception, 331; and philosophy, ers vs. holists, 271-72, 278-79; and logical
323-24, 329, 330, 336; and psychology, calculus, 320; and mental processes, 40;
130, 323-25, 327, 328, 329, 331, 336; and model of, 263; and neural commitment,
Zenon Pylyshyn's attack of Stephen Kos- 278; and neural functioning, 265-67; and
slyn, 332-34; and quasi-pictorial, 327, 331, perception, 10, 307; and phonology, 201;
334; and surface representations, 327; and plasticity of, 276-78; record of impulses,
three-dimensional objects, 330; and the 30; representational level, 39; and sensory
Wiirzburg School, 324 processing, 278; and specificity of func-
mentalism, 66, 194 tion, 273-74, 276, 284; and vision, 321; and
mentality, primitive vs. modem, 232-33 visual cortex, 273-74; and visual process-
Mental Models (Johnson-Laird), 367 ing systems, 319; see also neuroscience
mental models, 332, 364-65, 367-70, 384; neural: bases, 261; cells, 299; networks, 144,
and syllogistic reasoning, 368 169-70, 321; zones, 261
mental representations, 6, 38--40, 128-30, neural functioning, 165-67; and F. J. Gall,
313, 330, 332-34, 368--69, 383-84, 391; 266-67; and the Greeks, 265-66; and local-
and cognitive science, 38--40, 383, 391; ization, 265
and philosophy, 50 neurobehaviorial analysis, and Karl Lashley,
mental structures, 383 13
Method of Serial Reproduction, 115 neurobiology, 388
methodological solipsism, 84 neurochemistry, 282
methodology, molar vs. molecular, 135 neuroembryology, 270
mind: and computers, 6, 10, 51, 57, 67, 82, neurology, 75
148, 166, 223-25, 238--39, 264, 320, 331, neuronal model, 18--19; and logic, 361; and
353; and logical empiricism, 67; primitive, Warren McCulloch, 18, 19
223-25, 232; rational, 57; see also brain neurons, and parallel processing, 318-22
420
Subject Index
neurophysiology, 13, 75, 76, 176, 272, 273- ence, 308, 310, 313, 315; and intentional-
74, 282, 307, 358; and David Marr, 298 ity, 316; and language, 306; in David
neuropsychology, 22-23,260, 272; and Don- Marr's sketches, 301, 303, 307-8, 311; mel-
ald Hebb, 26; and holism, 269 ody, 111; and mental representations, 313,
neuroscience, 54, 260-88; and ablation, 261, 315; movement, 111; and nature of the
273; and behavior, 261, 262, 265; and brain, 300; and nerve-cell sensitivity, 296;
birds, 280-82; and cognitive science, 42, and parallel processing, 318-22; and per-
44--45, 286; and the Hixon Symposium, ceptual array, 311-12; and perceptual psy-
272-73; and holography, 282-84; and lan- chology, 309; and philosophy, 295-96;
guage, 287; and learning, 261-63; and le- principles of, 112, 114; proximity, 114; and
sion technique, 261; and localization, 261, psychology,24-25,295,296,300,302,309,
263, 265, 266, 267-71; and mental imagery, 310, 315; questions of, 295-96; and recog-
330; and molar perspective, 274-78; and nition, 298,304,308,311-12, 321-22;and
rationalism, 379; and reductionism, 261, stereopsis, 302; and symmetry, 114; theory
263, 264, 268, 269, 282, 285, 286; see also of, 286, 307; and top-down knowledge,
holism, localization, nervous system, neu- 304, 308, 318; visual, 131, 169-71, 219,
ral functioning 271, 297, 299; and visual processing,
New Science (Sdmza NOUR), 4, 5 301-5, 308, 319
New York Times, 110 PERCEPTRON, 169
Nobel Prize, 273, 275 phase sequences, 271
phase structure, 187-88, 211, 218; grammars,
186
phenomenology, 286
philologists, 198
philosophy, 31, 49-88, 136, 215-16, 378, 379,
Oedipus myth, 239, 240 381-82, 389-90; agenda, 53; and body, 51;
organizing schemas, 125-6 categories of, 340, 341, 351-5; of cognitive
Origin of Spedes, The, (Darwin), 392 science, 43, 44, 87-88, 389; and computers,
41, 78; critics of, 71; confrontation of Pop-
per and Wittgenstein, 76-77; and empiri-
cism, 52, 54, 60, 86, 206; foundational, 56-
59; and functionalism, 53; and human
thought, 50; and inference, 62; and infor-
paradigms, 130 mation processing, 51; and intentional sys-
parallel-processing systems, 318-22; and ar- tems, 79-81; and introspection, 50, 61, 73;
tificial intelligence, 319; and computers, and knowledge, 52; and language, 62-63,
319, 320-21; and the Establishment, 321; 212, 215-16, 217-20; and logic, 60-61, 360;
and Gibsonians, 321; and MPPS, 319; and and logical empiricism, 60-71; and man as
nervous system, 319 rational being, 360, 369, 378; and mathe-
Paris Neurological Society, 268 matics, 60-61; and mental imagery, 50,
parole, 196, 199 323-24, 329, 330, 336; and mental repre-
Particular Program, see Alfred P. Sloan Foun- sentations, 50; and the mind, 51; mind-
dation body problem, 31, 51, 78-79; and percep-
perception, 51, 100-1, 111-12, 131, 132, 164, tion, 51, 295; and physics, 87; and laws of
169, 170, 294, 295-322, 331, 358; afford- probability, 373-74; and psychology, 61;
ances, 310, 312, 315; and algorithm, 299- puzzles, 76; and rationalists, 53, 60, 86; ra-
300; and artificial intelligence, 169-71, 298, tionalists vs. empiricists, 8; and scientific
304-5, 306, 319; and bottom-up analysis, investigation, 53; and sensory experiences,
305, 312; brain, 112, 307; and categoriza- 52; and systemizers, 77; and the Vienna
tion, 354-55, 358, 359; and color naming, Circle, 63-65; see also empiricism, episte-
349-50; and computational theory, 299; in mology, foundational philosophy; logical
computers, 297-98, 300, 307; and connec- empiricism
tion with brain and computer, 306; direct, Philosophy and the Mirror of Nature (Rorty), 71,
304, 311; ecological approach, 314-15, 316, 76,85
321; and effectivities, 315; and empiricism, phonemes, 26, 200-1, 236; see 11/so utterances
55, 309; Establishment, 312-18, 319, 321; phonology, 188, 200-2, 206, 207-8; see also
Fodor-Pylyshyn view, 312-13; in Gestalt Prague School
psychology, 111-12, 276, 296; goals of, phrase-structure grammar, 186, 187, 188
302; and the Greeks, 296; and illusions, phrenology, 266
254; and implementation, 299; and infer- physicalism, and logical empiricism, 63
421
Subject Index
physical symbol system, 150 quasi-pictorial, 327, 331, 334, 336
physics, and psychologists, 102, 310
physiology, 103; of color vision, 347
plasticity, 262, 276-78, 284; and brain dam-
age, 276; Roger Sperry's results, 276
positivists, 65 Rand Corporation, 25, 145-46
pragmatics, 220 Ratio Club, 25, 26
Prague School, 200-2, 235-36; phonology, rationalism, 8, 53, 81, 86, 291, 317, 360, 363;
200-2 and molecular analysis, 97; and the Vienna
Primitive Culture (Tyler), 227 Circle, 65
Prindpia Mathemafica (Whitehead and Rus- rationality, 360-80, 399; and artificial intelli-
sell), 143, 145, 146, 147 gence, 379; in children, 377; and cognition,
probability, 372, 373, 375-76; and Bayes' 370-79; and computers, 379-80; empirical
law, 374-76 work in, 361; and the Greeks, 360; and
problem solving, in Gestalt psychology, 113 linguistics, 379; and neuroscience, 379; and
procedural representation, 161-62 philosophy, 360, 369, 373-79; and prefer-
process: encoding,131; execution, 131; match, ences, 373; and laws of probability, 372,
131; performance, 131; retrieval, 131 373, 374; and problem solving, 361; and
production system, 131, 150; in Newell- psychological paradigms, 377; and repre-
Simon's model, 131 sentativeness, 372; and laws of statistics,
prototype, 346 374; see also logic, mental models, reasoning
psycholinguistics, 214-15, 217, 219, 220, 379 reasoning, 292, 361-67, 367-70; set also logic,
psychology,10-11,24,25,32,54,69,74,89- rationality
137, 192-93, 194, 198, 201, 215-16, 286, reductionism, 261, 262-65, 268, 269, 282,
310, 324-25, 331, 336-39; and anthropol- 285, 286
ogy, 230, 254; and artificial intelligence, religions, 229
135, 137, 180-81; behaviorism, 109-11, representations: declarative, 161-62; mental,
119, 130, 135, 206, 260; Broadbent's flow 82-83, 330; primal sketch, 301, 303, 304,
chart, 92-93; and categories, 341, 350, 354; 308; procedural, 161-62; 3-D sketch, 301,
and classification, 93-95; and cognitive 303, 304; 2%-D sketch, 301-2, 303; su also
science, 136; and computer models, 95, mental representations
328, 329; and computers, 118-24; develop- representativeness, 372
mental, 116-18, 370; ecological, 315; edu- robots, 141, 173
cational, 370; empirical, 101, 103; and epis- Rockefeller Foundation, 138
temology, 74-75, 85; and foundational Russian Scandal, 115
philosophy, 59; and functionalism, 107-9;
funding in, 25; future of, 137; history of,
99; inference, 100; and information proc-
essing, 29, 90, 91, 119-21, 130-32, 134,
311; and information theory, 95; and in- schemas, 58--59, 116, 118, 126-27, 130, 163,
trospection, 101, 102-3, 107, 119; and lan- 317, 383
guage, 104; and learning theories, 94-95; Science Research Council (Great Britain),
limitations of processing, 90, 95; and lin- 164
guistics, 192-93, 201; and memory, 121- script, 165
24, 126-28, 131; and mental imagery, 323- semantics, 84, 188,191, 205,206,207-8,209,
25, 327, 328, 329, 331, 336-39; and mental 213, 215, 216, 251; generative, 209-10; in-
representations, 95, 128-30, 337; method- terpretive, 209; levels of analysis, 209; and
ology, 96-98, 135, 136; and modular anal- syntax, 212
ysis, 132-33; and molecular analysis, 119; semiotics, 245
and nee-grammarians, 198; and neuro- sensory experiences, 52, 58
science, 278; and optics, 310; and percep- sentences, 63-64, 69-70, 124-25, 182-85;
tion, 24, 25, 100-1, 295, 296, 311; percep- grammatical, 186; kernel, 187, 207; proper-
tual, 134, 309; and philosophy, 61; and ties of, 182-85
physics, 102, 310; and physiology, 103; of SHRDLU, 158-60, 164, 165, 170, .2.97
preferences, 373; and science, 91; scientific, Sloan Foundation, see Alfred P. Sloan Foun-
98-105; and self, 96; and top-down analy- dation
sis, 124-28, 135; see also cognitive psychol- social reality, 235
ogy, Gestalt psychology Society of Fellows (Harvard University),
psychophysics, 96 25-26
psychophysiology, 75 Society of Minds, 166
422
Subject Index
split brain, 275 TOTE (Test-Operate-Test-Exit), 33, 264
S-structure, 218 Tradalus (Wittgenstein), 62
Standard Theory, in linguistics, 207-8 transformational: anthropology, 242.-43; com-
State of the Art Report (SOAP), 36-37 ponent, 208, 210, 217; grammar, 186-88,
stereopsis, 302 215, 216, 218; linguistics, 177, 191, 208,
stereoscopic vision, 300; computation of, 2.51; power, 210; psychological reality,
300 214; revolution, 204, 207; rules, 210-11;
stories, 125-27; and information processing, structures, 210
125-27; schemas, 130 Turing Machine, 17, 18, 31, 79, 144,
structural anthropology, 235, 236-44, 258; 172
and kinship, 236-37, 239; and linguistics,
236, 237
structuralism, 208, 213, 236-44; German,
113; and Gestalt psychology, 113; and
linguistics, 191, 195, 201, 202, 208, 213,
236 Umwt!lt, 31
Structure of LanguJlge, The (Fodor and Katz), 34 unconscious inference, 100
Study of Thinking, A (Bruner, Goodnow, and understanding system, 168
Austin), 93, 95 universal grammar (U.G.), 211-12; subsys-
superordinate, 346 tems of, 211
surface structure, 208, 209, 211 utterances, 69-70, 160, 184, 203; and philos-
syllogisms, 363-64,365-66, 367-70; and men- ophy, 69-70
tal models, 367-70
"Symbolic Analysis of Relay and Switching
Circuits, A" (Shannon), 144
symbolic processes, 243-44
Symposium for Information Theory, MIT,
28 verification, 63
Syntactic Structures (Chomsky), 182, 184, 207, Vienna Circle, 63-65, 70, 73, 76, 292, 351;
213; responses to, 189-90 goals of, 63; members of, 63; Nelson
syntax, 183-84, 185, 190, 191, 193, 201, 203, Goodman's views of, 64-65
206-7, 209, 210, 213, 220, 222; analysis of, vision: and David Marr, 2.98; and neurons,
203, 222; autonomous, 220; and finite- 300; and steps of processing, 301-4;
state grammar, 185, 188; and logical em- stereoscopic, 300
piricism, 64; and perception, 306; and se- visual cortex, 273-74
mantics, 212; theory, 218; and trans- visual perception, see perception
formations, 188 visual processing, 301-5, 308, 319
visual systems, 270
Third Texas Conference on Problems of lin-

guistic Analysis of English, 194 What Computers Can 'f Do (Dreyfus), 163
thought, 56, 223-25, 23!h39; in children, Wurzburg School, 109, 113, 118, 126, 127;
117; and continental tradition, 57; and phi- and imagery, 324; members of, 105-6; re-
losophy, 50; pre-logic, 223-24; in primi- search of, 105-7
tive people, 223-25, 232, 257; see also mind
top-down approach: and artificial intelli-
gence, 165--66; and psychology, 97
Torres Expedition, see Cambridge Anthropo-
logical Expedition Xerox Corporation, 178
423

Pub The Minds New Science A History of The Cognitive R

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pub The Minds New Science A History of The Cognitive R

Uploaded by

Copyright:

Available Formats

The Mind's

The Quest for Mind (1973; second edition, 1981)

Basic Books, Inc., Publishers I New York

Library of Congress Cataloging in Publication Data

Copyright © 1985 by Howard Gardner

2 Laying the Foundation for Cognitive Science 10

3 Cognitive Science: The First Decades 28

4 Reason, Experience, and the Status of Philosophy 49

5 Psychology: The Wedding of Methods to

6 Artificial Intelligence: The Expert Tool 138

7 Linguistics: The Search for Autonomy 182

8 Anthropology: Beyond the Individual Case 223

9 Neuroscience: The Flirtation with Reductionism 260

10 Perceiving the World 295

11 Mental Imagery: A Figment of the Imagination? 323

12 A World Categorized 340

13 How Rational a Being? 360

14 Conclusion: The Computational Paradox and the

In the mid-1970s, I began to hear the term cognilive science. As a psychologist

Mara Krechevsky, my current research assistant, has been invaluable in

The safest general characterization of the European philo-

The Greek Agenda

The centerpiece of the dialogue features many questions and re-

Definition and Scope of Cognitive Science

Purpose and Plan of This Book

The Hixon Symposium and the Challenge to Behaviorism

In September of 1948 on the campus of the California Institute of Technol-

dividuals acting as they do because of their own ideas and intentions, or

dressed them, Lashley was adopting a radical position. Scientists concerned

A Critical Moment in Scientific History

The scholars in attendance at the Hixon Symposium stood at a critical

Key Theoretical Inputs to Cognitive Science

Mathematics and Computation

The Neuronal Model

The Cybernetic Synthesis

The central nervous system no longer appears as a self-contained organ,

The parallels to Lashley's ideas about neural organization-and the chal-

of information. Only very recently have cognitive scientists begun to

Catalytic Encounters and Influential Writings

By the 1940s, then, the principal intellectual capital on which cognitive

In addition to these many face-to-face encounters among those con-

day, Ashby deliberately avoided talking of anything like consciousness or

Miller's testimony is corroborated by other witnesses. From the ranks

This impressive congruence stresses a few seminal publications, ema-

The 1960s: Picking Up Steam

None of [these programs] does even remote justice to the complexity of

After Neisser, it was possible to buy the cognitive science approach in

We may be at the start of a major intellectual adventure: somewhere compara-

The Sloan Initiative

Connections among the Cognitive Sciences

In my view, the authors of the SOAP document made a serious effort

Key Features of Cognitive Science

In my own work I have found it useful to distinguish five features or

ble to posit multiple forms of mental representation and that it is impossi-

De-Emphasis on Affect, Context, Culture, and History

to so define and investigate problems that an adequate account can be

Belief in Interdisciplinary Studies

Rootedness in Classical Philosophical Problems

hunch: that human thought would tum out to resemble in significant

There is no department of knowledge in which so little

In this chapter I begin my survey of the cognitive sciences by examining

topics upon which empirically oriented cognitive scientists are working

And what I believe to be more important here is that I find in myself an